WO2024253400A1 - Apparatus and system capable of correcting position error through real-time analysis and comparison between vision data and lidar data for slam technology implementation - Google Patents

Abstract

The present invention relates to an apparatus and a system capable of correcting a position error for SLAM technology implementation. For example, disclosed is an apparatus comprising: a DB module for matching and storing pre-generated multiple pieces of vision data and multiple pieces of standard position information, wherein the data is stored in multiple position groups which are grouped, respectively, according to configuration areas having a predetermined coordinate range; a sensor module for calculating estimated position information of an autonomous driving robot via LiDAR and odometry, and specifying one position group among the multiple position groups according to the calculated estimated position information; a vision module for providing a vision recognition result via a vision recognition process for a driving image of the autonomous driving robot; a position information extraction module for extracting standard position information matched to vision data that is most similar to the vision recognition result from among vision data belonging to the position group specified via the sensor module; and a position information reconfiguration module for comparing the standard position information extracted via the position information extraction module with the estimated position information calculated via the sensor module, and reconfiguring and correcting position information of the autonomous driving robot according to a comparison result.

Description

Device and system capable of position error correction through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology

An embodiment of the present invention relates to a device and system capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology.

Autonomous driving refers to driving to a destination by making decisions on your own without the intervention of a robot or user. To do this, you need a 2D or 3D map of the environment you want to drive in. Since a 2D map is made on a flat surface, obstacles that are not recognized by the 2D sensor are not mapped.

The technology that allows a robot to create a map of a space using sensors such as LiDAR and cameras while estimating its own location so that it can safely reach its destination autonomously in a space where there is no environmental information is called SLAM (Simultaneous Localization and Mapping).

Among SLAM technologies, Vision-Based SLAM (vSLAM) is a method that can create a 3D map using an image sensor (camera). However, since cameras have a narrow field of view, mapping using only cameras may result in missing data.

At this time, the camera can be largely divided into three types: 'Monocular', 'Stereo', and 'RGB-D'. Here, the 'Monocular Camera' uses only one camera sensor. Therefore, it has the advantage of being very simple in structure and cheap, but since it loses depth or distance information, 'Monocular SLAM' can predict 'motion', 'distance', 'size' of 'object', etc. through continuous images according to the movement of the camera.

vSLAM's framework can be composed of 'Sensor Data Acquisition', 'Visual Odometry (VO)', 'Backend Filtering/Optimization', 'Loop Closing', and 'Reconstruction'.

'Sensor Data Acquisition' is the process of obtaining and preprocessing camera images. 'Motor encoder' and 'IMU sensor values' can be obtained from a moving robot, and the 'synchronization' process can also be performed.

'Visual Odometry (VO)' can estimate how the camera moved between adjacent frames and create a rough local map.

'Backend Filtering/Optimization' can receive camera poses and 'loop closing' results at different time steps and apply an optimization algorithm to create an optimized trajectory and map.

'Loop Closing' can provide information (whether the robot has returned to the previous position) to 'optimization' in the 'backend' by determining whether the robot has returned to the previous position to reduce the accumulated drift. Here, the drift problem refers to the problem that occurs when errors continuously occur in 'Visual Odometry' and accumulate.

'Reconstruction' can construct a map according to the robot's task based on the estimated camera trajectory.

As described above, 'Visual Odometry' estimates the camera movement between adjacent image frames and connects them based on time steps. Since the error occurring at each time step is applied to the next 'Visual Odometry', errors continue to accumulate, which is called the 'accumulated drift problem'.

'Loop Closing' is a method to solve the 'accumulated drift problem'. When the robot moves to a certain extent and returns to its original position, the estimated result may not return to the origin due to the error at that time. If there is a way to tell that the robot has returned to the origin, the error can be removed by pulling the estimated position information back to the origin.

One way to make the robot recognize the origin is to place a marker at the starting point, and when the robot recognizes the marker, it can be known that the robot has returned to the starting point.

However, since markers are greatly affected by the environment, it is preferable to obtain 'Loop Closing' using the image itself rather than the marker.

Prior art of the present invention is Patent Publication No. 10-2022-0152451 (publication date: November 16, 2022).

An embodiment of the present invention provides a SLAM position error correction device and a SLAM position error correction system that are linked with a cloud device, which configures a database by matching images and coordinate information based on visual information, performs a similarity test between images acquired in real time and the database through visual loop detection, estimates position information using visual information, and then readjusts a position estimated by LiDAR and odometry using the estimated information.

According to one embodiment of the present invention, a device capable of correcting position errors through real-time analysis and comparison between vision data and LiDAR data for implementing SLAM (Simultaneous Localization And Map-Building) technology includes: a DB module that matches and stores a plurality of vision data and a plurality of standard location information generated in advance, and stores them as a plurality of location groups each grouped according to a set area having a certain coordinate range; a sensor module that calculates estimated location information of an autonomous driving robot through LiDAR and odometry and specifies one of the plurality of location groups according to the calculated estimated location information; a vision module that provides a vision recognition result through a vision recognition process for a driving image of the autonomous driving robot; a location information extraction module that extracts standard location information that matches vision data most similar to the vision recognition result among vision data belonging to the location group specified through the sensor module; And it includes a location information reset module that compares the standard location information extracted through the location information extraction module and the estimated location information calculated through the sensor module, and resets and corrects the location information of the autonomous driving robot according to the comparison result.

A system capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM (Simultaneous Localization And Map-Building) technology according to another embodiment of the present invention comprises: an autonomous driving robot including a sensor module and a vision module; And a DB module, a location information extraction module, and a location information reset module are included, and a cloud server connected to the autonomous driving robot through an Internet communication network, wherein the DB module matches a plurality of vision data and a plurality of standard location information generated in advance and stores them, and stores them as a plurality of location groups each grouped according to a set area having a certain coordinate range, the sensor module calculates estimated location information of the autonomous driving robot through LiDAR and Odometry, and specifies one location group among the plurality of location groups according to the calculated estimated location information, the vision module provides a vision recognition result through a vision recognition process for a driving image of the autonomous driving robot, the location information extraction module extracts standard location information that matches vision data most similar to the vision recognition result among vision data belonging to the location group specified through the sensor module, and the location information reset module compares the standard location information extracted through the location information extraction module and the estimated location information calculated through the sensor module, and resets and corrects the location information of the autonomous driving robot according to the comparison result.

According to another embodiment of the present invention, a system capable of correcting position errors through real-time analysis and comparison between vision data and LiDAR data for implementing SLAM (Simultaneous Localization And Map-Building) technology comprises: a DB module that matches and stores a plurality of vision data and a plurality of standard location information generated in advance, and stores them as a plurality of location groups each grouped according to a set area having a certain coordinate range; a sensor module that calculates estimated location information of an autonomous driving robot through LiDAR and odometry and specifies one of the plurality of location groups according to the calculated estimated location information; a vision module that provides a vision recognition result through a vision recognition process for a driving image of the autonomous driving robot; a location information extraction module that extracts standard location information that matches vision data that is most similar to the vision recognition result among vision data belonging to the location group specified through the sensor module; An autonomous driving robot including a location information reset module that compares the standard location information extracted through the location information extraction module with the estimated location information produced through the sensor module, and resets and corrects the location information of the autonomous driving robot based on the comparison result; and a DB monitoring module that monitors data stored in the DB module and generates DB monitoring data; and a DB backup module that is synchronized with the DB module and backs up the data stored in the DB module; and a DB modification module that determines and deletes data that is not necessary for the DB module based on the DB monitoring data provided from the DB monitoring module, and includes a cloud server connected to the autonomous driving robot through an Internet communication network.

A system capable of correcting position errors through real-time analysis and comparison between vision data and LiDAR data for implementing SLAM (Simultaneous Localization And Map-Building) technology according to another embodiment of the present invention comprises: a first DB module which matches and stores a plurality of vision data and a plurality of standard location information generated in advance, and stores them as a plurality of location groups each grouped according to a set area having a certain coordinate range; a sensor module which calculates estimated location information of an autonomous driving robot through LiDAR and odometry and specifies one of the plurality of location groups according to the calculated estimated location information; a vision module which provides a vision recognition result through a vision recognition process for a driving image of the autonomous driving robot; a first location information extraction module which extracts standard location information which matches vision data most similar to the vision recognition result among vision data belonging to the location group specified through the sensor module; And an autonomous driving robot including a location information reset module that compares the standard location information extracted through the first location information extraction module and the estimated location information produced through the sensor module, and resets and corrects the location information of the autonomous driving robot according to the comparison result; And a second DB module that matches and stores a plurality of vision data and a plurality of standard location information generated in advance, and stores them as a plurality of location groups each grouped according to a set area having a certain coordinate range; And a second location information extraction module that extracts standard location information that matches the vision data most similar to the vision recognition result among the vision data belonging to the location group specified through the sensor module when the standard location information is not extracted through the first location information extraction module, and provides the extracted standard location information to the first location information extraction module, and includes a cloud server connected to the autonomous driving robot through an Internet communication network.

According to the present invention, a SLAM position error correction device and a SLAM position error correction system that are linked with a cloud device can be provided, which configures a database by matching images and coordinate information based on visual information, performs a similarity test between images acquired in real time and the database through visual loop detection, estimates position information using visual information, and then readjusts a position estimated by LiDAR and odometry using the estimated information.

FIG. 1 is a drawing showing the overall configuration and configuration form of a SLAM position error correction device applied to an autonomous driving robot according to the first embodiment of the present invention.

FIG. 2 is a block diagram showing the overall configuration and configuration relationship of a SLAM position error correction device according to the first embodiment of the present invention.

FIG. 3 is a drawing illustrating the overall operation process of a SLAM position error correction device according to the first embodiment of the present invention.

Figure 4 is a block diagram showing the configuration of a DB module according to the first embodiment of the present invention.

FIG. 5 is a graph showing an example of selecting a location group according to the first embodiment of the present invention.

Figure 6 is a block diagram showing the configuration of a vision module according to the first embodiment of the present invention.

Figure 7 is a block diagram showing the configuration of a location information extraction module according to the first embodiment of the present invention.

FIG. 8 is a drawing illustrating an operation process according to position readjustment of a position information reset module according to the first embodiment of the present invention.

FIG. 9 is a diagram illustrating the overall operation process of the DB update module according to the first embodiment of the present invention.

FIG. 10 is a drawing showing the overall configuration and configuration form of a SLAM position error correction system according to the second embodiment of the present invention.

Figure 11 is a block diagram showing the configuration of a vision module according to the second embodiment of the present invention.

Figure 12 is a block diagram showing the configuration of a DB module according to the second embodiment of the present invention.

Figure 13 is a block diagram showing the configuration of a location information extraction module according to the second embodiment of the present invention.

FIG. 14 is a drawing showing the overall configuration and configuration form of a SLAM position error correction system according to the third embodiment of the present invention.

Figure 15 is a block diagram showing the configuration of a DB module according to the third embodiment of the present invention.

Figure 16 is a block diagram showing the configuration of a vision module according to a third embodiment of the present invention.

Figure 17 is a block diagram showing the configuration of a location information extraction module according to the third embodiment of the present invention.

FIG. 18 is a drawing showing the overall configuration and configuration form of a SLAM position error correction system according to the fourth embodiment of the present invention.

Figure 19 is a block diagram showing the configuration of a DB module according to the fourth embodiment of the present invention.

Figure 20 is a block diagram showing the configuration of a vision module according to the fourth embodiment of the present invention.

Figure 21 is a block diagram showing the configuration of a location information extraction module according to the fourth embodiment of the present invention.

FIG. 1 is a diagram showing the overall configuration and configuration form of a SLAM position error correction device applied to an autonomous driving robot according to the first embodiment of the present invention, FIG. 2 is a block diagram showing the overall configuration and configuration relationship of a SLAM position error correction device according to the first embodiment of the present invention, FIG. 3 is a diagram illustrating the overall operation process of the SLAM position error correction device according to the first embodiment of the present invention, FIG. 4 is a block diagram showing the configuration of a DB module according to the first embodiment of the present invention, FIG. 5 is a graph showing an example of selecting a position group according to the first embodiment of the present invention, FIG. 6 is a block diagram showing the configuration of a vision module according to the first embodiment of the present invention, FIG. 7 is a block diagram showing the configuration of a position information extraction module according to the first embodiment of the present invention, FIG. 8 is a diagram illustrating the operation process according to position readjustment of a position information reset module according to the first embodiment of the present invention, and FIG. 9 is a diagram illustrating the overall operation process of a DB update module according to the first embodiment of the present invention.

Referring to FIGS. 1 and 2, a SLAM position error correction device (1000) applied to an autonomous driving robot (10) may include at least one of a DB module (1100), a sensor module (1200), a vision module (1300), a position information extraction module (1400), a position information reset module (1500), and a DB update module (1600).

The above DB module (1100) can store a plurality of pre-generated vision data and a plurality of standard location information by matching them with each other, and can store them as a plurality of location groups each grouped according to a set area having a certain coordinate range.

Such DB module (1100) may include at least one of a location group creation unit (1110) and a location group management unit (1120), as illustrated in FIG. 4.

The above location group generation unit (1110) matches the DB vector data extracted from each vision data (image of a driving video) and the standard location information representing the coordinate information where each vision data is generated to generate each data set, and groups the generated multiple data sets according to a set area having a certain coordinate range to generate multiple location groups. Here, the set area means a range for selecting a location group for performing a similarity search at the location of the autonomous driving robot (10), and standard location information that is divided and fixed at a certain interval is defined in one location group.

For example, in the case of a setting area 'S1' (corresponding to one location group) consisting of a certain coordinate range as illustrated in Fig. 5, since the location where the autonomous driving robot (10) is displayed is in the center of the group, a similarity comparison search can be performed only in that group, and in the case of 'S2', since the location where the autonomous driving robot (10) is displayed is located at the boundary of the group, it means that the surrounding groups must also be searched, and accordingly, the group where a similarity comparison search is to be performed can be selected relatively more widely, such as 'S2'.

As described above, in one location group, a number of standard location information (location), i.e., standard coordinate data, are defined at regular intervals, and an image matching each standard coordinate is defined as vector data.

That is, since images captured within a certain area are similar to each other (images captured of the surrounding environment have similar characteristics), such similar images can be grouped together, and the vector data for each image within each group and the coordinate data when each image was captured can be matched to form a single data set, and multiple data sets (vector data, standard location information) corresponding to each image can be defined or stored within a single location group.

The above location group management unit (1120) can store a data set of a location group created through the location group creation unit (1110), and when a data update command and data to be updated are input by the DB update module (1600), it can modify previously stored data.

The above sensor module (1200) calculates estimated location information (information estimating the location of the robot itself) of the autonomous driving robot (10) through LiDAR and Odometry systems, and can specify one location group among a number of location groups defined in the DB module (1100) based on the calculated estimated location information.

LiDAR can detect and measure the distance to a target object by emitting a 360-degree laser pulse and calculating the time it takes for the light to reflect off surrounding objects such as walls or obstacles and return to recognize the surrounding environment. Odometry measures the rotational speed through an encoder and the inclination, etc., using an IMU sensor to measure the position of a moving object, thereby finding its own position on the map, that is, the position of the autonomous driving robot (10) (estimated position information), and driving.

The above sensor module (1200) can compare the estimated location information of the autonomous driving robot (10) calculated through LiDAR and Odometry with the set area defined for each location group to determine the location group corresponding to the estimated location information of the autonomous driving robot (10).

As described above, since each location group has a defined location range, i.e., a coordinate range, the sensor module (1200) can check which location group the estimated location information of the autonomous driving robot (10) belongs to and output the estimated location information and the location group ID or identification information corresponding to the estimated location information.

The above vision module (1300) can provide a vision recognition result through a vision recognition process for a driving image of an autonomous driving robot (10).

To this end, the vision module (1300) may include at least one of a driving image generation unit (1310), a driving image acquisition unit (1320), and a vector data extraction unit (1330), as illustrated in FIG. 6.

The above driving image generation unit (1310) can generate driving images of the autonomous driving robot (10) through at least two cameras installed in the autonomous driving robot (10) and shooting from different directions or angles.

For example, to increase the accuracy of image similarity comparison, driving images of an autonomous driving robot (10) can be generated using a camera facing the ceiling and a camera facing the front, respectively.

The above driving image acquisition unit (1320) can acquire images of driving images of the autonomous driving robot (10) at regular intervals and assign an image ID to each acquired image.

The above vector data extraction unit (1330) can extract current vector data from each driving image acquired through the driving image acquisition unit (1320) and provide it as a vision recognition result.

The 'current vector data' and 'DB vector data' mentioned below are both composed of lines created by connecting points and are data that can be expressed as a mathematically calculated functional relationship, and are terms of the same concept. However, in order to help understand the invention, the vector data generated in the vision module (1300) is defined as 'current vector data', and the vector data stored in the DB module (1100) is defined as 'DB vector data' to distinguish similar terms.

In addition, the 'estimated location information' described above and the 'standard location information' mentioned below are the same in that they mean coordinate information indicating the location of the autonomous driving robot (10), but the 'estimated location information' is coordinate information calculated through the sensor module (1200), and the 'standard location information' can be distinguished as coordinate information that is matched with 'DB vector data' in the DB module (1100) and stored in advance.

The above location information extraction module (1400) can find the vision data (DB vector data) most similar to the vision recognition result (current vector data) among the vision data (DB vector data) belonging to a specific location group through the sensor module (1200), and extract standard location information matching the corresponding vision data (DB vector data).

To this end, the location information extraction module (1400) may include at least one of a data input unit (1410), a location group selection unit (1420), a similarity result calculation unit (1430), and a location information provision unit (1440), as illustrated in FIG. 7.

The above data input unit (1410) can receive current vector data from the vision module (1300) and input estimated location information and specific location group information (location group ID or identification information) from the sensor module (1200), respectively. Accordingly, information on the image currently being captured through the vision module (1300), location information recognized through the sensor module (1200), and information on the location group corresponding to the location information can be input, respectively.

The above location group selection unit (1420) can select a location group stored in the DB module (1100) according to the location group information (location group ID or identification information) input through the data input unit (1410). In this way, instead of performing a comparison process on the entire DB image to find an image similar to the image recognized through the vision module (1300), by specifying a group belonging to the location recognized through the sensor module (1200), candidate images to be compared with the image recognized by the vision module (1300) can be specified, thereby reducing the number of comparison targets. Accordingly, since the number of comparisons is reduced, the time required for the similarity comparison process can be reduced, and errors such as cases where images are similar but located in different places can be eliminated, thereby increasing accuracy.

The above similarity result generating unit (1430) can perform a cosine similarity measurement between 'DB vector data' belonging to a location group selected through a location group selecting unit (1420) and 'current vector data' input through a data input unit (1410) and calculate a score accordingly.

The above location information providing unit (1440) checks whether the highest similarity score among the similarity scores calculated through the similarity result calculating unit (1430) exceeds a preset threshold score, and if the highest similarity score exceeds the threshold score, extracts and provides standard location information that matches the DB vector data that received the highest similarity score.

For example, if a cosine similarity comparison is performed between 'current vector data K' and DB vector data '1~10' included in 'location group 2', and the similarity score for DB vector data '2' is measured to be the highest, the similarity score of DB vector data '2' is compared with a threshold score, and if it is confirmed that the similarity score of DB vector data '2' exceeds the threshold score, the standard location information '2' matching DB vector data '2' can be extracted.

The above location information reset module (1500) can compare the standard location information extracted through the location information extraction module (1400) and the estimated location information calculated through the sensor module (1200), and reset the location information of the autonomous driving robot (10) based on the comparison result.

More specifically, the location information reset module (1500) compares the estimated location information produced by the sensor module (1200) with the standard location information provided by the location information provider (1440) to determine whether they match, and if they do not match, as shown in FIG. 8, it determines that there is an abnormality in the driving state of the autonomous driving robot (10), resets the LiDAR and Odometry of the sensor module (1200) while the autonomous driving robot (10) is temporarily stopped, and then sets the standard location information provided by the location information provider (1440) as the initial location information of the LiDAR and Odometry to correct the location error.

This location information reset module (1500) performs a similarity check between an image acquired in real time through visual loop detection and a pre-built database, and can readjust the initial location information from LiDAR and Odometry based on the location information estimated using visual information when there is an abnormality in the driving status of the autonomous driving robot (10).

Here, position readjustment refers to a process of resetting the current position information of the autonomous driving robot (10) by resetting the odometry and assigning initial position information after temporarily stopping the autonomous driving robot (10) when it is determined that the driving state of the autonomous driving robot (10) is incorrect, as shown in FIG. 8. In this way, when assigning an initial position (initial pose) in the existing SLAM algorithm, an algorithm for estimating the current position information is used to reset the position information estimated using visual information to the initial position information when an abnormality in the driving state of the autonomous driving robot (10) is detected, thereby correcting the error in the current position information.

The above DB update module (1600) adds the image of the current vector data having the highest similarity score to the candidate DB if the highest similarity score does not exceed a threshold, and then performs a similarity comparison with vector data previously added to the candidate DB. If the image exceeds the threshold, it performs an up-count on the image ID of the corresponding vector data. If the image does not exceed the threshold, it assigns a new image ID and adds it to the candidate DB. However, if the cumulative counter count for the corresponding image ID exceeds a preset threshold count by repeatedly performing the cosine similarity score calculation process and the threshold score comparison process, the DB module (1100) can update the data set of the DB module (1100) by additionally registering the average value of the current vector data having the highest similarity score and the estimated location information corresponding to the current vector data.

That is, if the highest similarity score does not exceed the reference score, the image of the current vector data with the highest similarity score is added to the candidate DB, and the vector data added to the candidate DB is compared for similarity with the vector data existing in the existing candidate DB. If there is a similar vector (i.e., if it exceeds the reference score), an upcount is performed on the image ID of the corresponding vector data and estimated location information is added. If there is no similar vector data, a new ID is assigned to the image of the corresponding vector data and stored in the candidate DB.

By repeating this process, when the accumulated counter count exceeds a preset standard count, the average value of the corresponding data and the estimated location information corresponding to the corresponding data is additionally registered in the DB module (1100), thereby updating the data set of the DB module (1100), and the corresponding data can be removed from the candidate DB.

In this way, a process of adding data judged as new data through a similarity comparison with the DB module (1100) to the candidate DB and a process of finding reliable new vector data through a similarity comparison in the candidate DB when vector data is added to the candidate DB (i.e., a method using the accumulated counter count) can be performed.

As the autonomous driving robot (10) drives and takes pictures of the inside of a building, the inside of the building may change over time, so the image (i.e., vector data) stored in the DB module (1100) must also be updated accordingly. Accordingly, when the DB update module (1600) compares the image provided through the vision module (1300) with the image stored in the DB module (1100), if a similar image is not found (i.e., if the highest similarity score does not exceed the reference score), vector data and coordinate information for the corresponding image can be additionally registered in the DB module (1100).

However, since determining the additional registration target through a single judgment process may rather contaminate the DB, if the above condition (the number of accumulated counters is greater than the preset standard number of times) is satisfied N or more times (e.g. 3 or more times), the corresponding vector data (current N pieces of vector data) can be added to the DB module (1100), and at this time, the average value of the N pieces of estimated location information provided by the sensor module (1200) when the current vector data is acquired from the vision module (1300) can be applied as the location information.

FIG. 10 is a drawing showing the overall configuration and configuration form of a SLAM position error correction system according to the second embodiment of the present invention, FIG. 11 is a block diagram showing the configuration of a vision module according to the second embodiment of the present invention, FIG. 12 is a block diagram showing the configuration of a DB module according to the second embodiment of the present invention, and FIG. 13 is a block diagram showing the configuration of a position information extraction module according to the second embodiment of the present invention.

Referring to FIG. 10, the SLAM position error correction system (2000) according to the second embodiment may include at least one of an autonomous driving robot (2100) and a cloud server (2200).

The above autonomous driving robot (2100) may include at least one of a sensor module (2110) and a vision module (2120), and the cloud server (2200) may include at least one of a DB module (2210), a location information extraction module (2220), a location information reset module (2230), and a DB update module (2240).

The above autonomous driving robot (2100) is connected to a cloud server (2200) through an Internet communication network and can perform an operation for position error correction by interworking with the cloud server (2200).

Below, each component is described in detail in the following order: a sensor module (2110) and a vision module (2120) configured in an autonomous driving robot (2100), and a DB module (2210), a location information extraction module (2220), a location information reset module (2230), and a DB update module (2240) configured in a cloud server (2200).

The above sensor module (2110) calculates estimated location information (information estimating the location of the robot itself) of the autonomous driving robot (2100) through LiDAR and Odometry systems, and can specify one location group among a number of location groups defined in the DB module (2210) based on the calculated estimated location information.

The above sensor module (2110) can compare the estimated location information of the autonomous driving robot (2100) calculated through LiDAR and Odometry with the set area defined for each location group to determine the location group corresponding to the estimated location information of the autonomous driving robot (2100).

As described above, since each location group has a defined location range, i.e., a coordinate range, the sensor module (2110) can check which location group the estimated location information of the autonomous driving robot (2100) belongs to and output the estimated location information and the location group ID or identification information corresponding to the estimated location information.

The above vision module (2120) can provide a vision recognition result through a vision recognition process for a driving image of an autonomous driving robot (2100).

To this end, the vision module (2120) may include at least one of a driving image generation unit (2121), a driving image acquisition unit (2122), and a vector data extraction unit (2123), as illustrated in FIG. 11.

The above driving image generation unit (2121) can generate driving images of the autonomous driving robot (2100) through at least two cameras installed in the autonomous driving robot (2100) and shooting images in different directions or angles.

For example, to increase the accuracy of image similarity comparison, driving images of an autonomous driving robot (2100) can be generated using a camera facing the ceiling and a camera facing the front, respectively.

The above driving image acquisition unit (2122) can acquire images of driving images of the autonomous driving robot (2100) at regular intervals and assign an image ID to each acquired image.

The above vector data extraction unit (2123) can extract current vector data from each driving image acquired through the driving image acquisition unit (2122) and provide it as a vision recognition result.

The above DB module (2210) can store a plurality of pre-generated vision data and a plurality of standard location information by matching them with each other, and can store them as a plurality of location groups each grouped according to a set area having a certain coordinate range.

Such DB module (2210) may include at least one of a location group creation unit (2211) and a location group management unit (2212), as illustrated in FIG. 12.

The above location group generation unit (2211) matches the DB vector data extracted from each vision data (image of a driving video) and the standard location information representing the coordinate information where each vision data is generated to generate each data set, and groups the generated multiple data sets according to a set area having a certain coordinate range to generate multiple location groups. Here, the set area means a range for selecting a location group for performing a similarity search at the location of the autonomous driving robot (10), and standard location information that is divided and fixed at a certain interval is defined in one location group.

For example, in the case of a setting area 'S1' (corresponding to one location group) consisting of a certain coordinate range as illustrated in Fig. 5, since the location where the autonomous driving robot (10) is displayed is in the center of the group, a similarity comparison search can be performed only in that group, and in the case of 'S2', since the location where the autonomous driving robot (10) is displayed is located at the boundary of the group, it means that the surrounding groups must also be searched, and accordingly, the group where a similarity comparison search is to be performed can be selected relatively more widely, such as 'S2'.

The above location group management unit (2212) can store a data set of a location group created through a location group creation unit (2211), and when a data update command and data to be updated are input by a DB update module (2240), it can modify previously stored data.

The above location information extraction module (2220) can find the vision data (DB vector data) most similar to the vision recognition result (current vector data) among the vision data (DB vector data) belonging to a specified location group through the sensor module (2110) and extract standard location information matching the vision data (DB vector data).

To this end, the location information extraction module (2220) may include at least one of a data input unit (2221), a location group selection unit (2222), a similarity result calculation unit (2123), and a location information provision unit (2224), as illustrated in FIG. 13.

The above data input unit (2221) can receive current vector data from the vision module (2120) and input estimated location information and specific location group information (location group ID or identification information) from the sensor module (2110), respectively. Accordingly, information on the image currently being captured through the vision module (2120), location information recognized through the sensor module (2110), and information on the location group corresponding to the location information can be input, respectively.

The above location group selection unit (2222) can select a location group stored in the DB module (2210) based on the location group information (location group ID or identification information) input through the data input unit (2221). The above similarity result calculation unit (2123) can perform a cosine similarity measurement between the 'DB vector data' belonging to the location group selected through the location group selection unit (2222) and the 'current vector data' input through the data input unit (2221) and calculate a score accordingly.

The above location information providing unit (2224) checks whether the highest similarity score among the similarity scores calculated through the similarity result calculating unit (2123) exceeds a preset threshold score, and if the highest similarity score exceeds the threshold score, extracts and provides standard location information that matches the DB vector data that received the highest similarity score.

The above location information reset module (2230) can compare the standard location information extracted through the location information extraction module (2220) and the estimated location information calculated through the sensor module (2110), and reset the location information of the autonomous driving robot (10) based on the comparison result.

More specifically, the location information reset module (2230) compares the estimated location information produced by the sensor module (2110) with the standard location information provided by the location information provider (2224) to determine whether they match, and if they do not match, determines that there is an abnormality in the driving status of the autonomous driving robot (10), resets the LiDAR and Odometry of the sensor module (2110) while the autonomous driving robot (10) is temporarily stopped, and then sets the standard location information provided by the location information provider (122s) as the initial location information of the LiDAR and Odometry to correct the location error.

This location information reset module (2230) performs a similarity check between an image acquired in real time through visual loop detection and a pre-built database, and can readjust the initial location information from LiDAR and Odometry based on the location information estimated using visual information when there is an abnormality in the driving status of the autonomous driving robot (10).

The above DB update module (2240) adds the image of the current vector data having the highest similarity score to the candidate DB if the highest similarity score does not exceed a threshold, and then performs a similarity comparison with vector data previously added to the candidate DB. If the similarity score exceeds the threshold, it performs an up-count on the image ID of the corresponding vector data. If the similarity score does not exceed the threshold, it assigns a new image ID and adds it to the candidate DB. However, if the cumulative counter count for the corresponding image ID exceeds a preset threshold count by repeatedly performing the cosine similarity score calculation process and the threshold score comparison process, the DB module (2210) can additionally register the average value of the current vector data having the highest similarity score and the estimated location information corresponding to the current vector data to the DB module (2210) to update the data set of the DB module (2210).

FIG. 14 is a drawing showing the overall configuration and configuration form of a SLAM position error correction system according to the third embodiment of the present invention, FIG. 15 is a block diagram showing the configuration of a DB module according to the third embodiment of the present invention, FIG. 16 is a block diagram showing the configuration of a vision module according to the third embodiment of the present invention, and FIG. 17 is a block diagram showing the configuration of a position information extraction module according to the third embodiment of the present invention.

Referring to FIG. 14, the SLAM position error correction system (3000) according to the third embodiment may include at least one of an autonomous driving robot (3100) and a cloud server (3200).

The above autonomous driving robot (3100) may include at least one of a DB module (3110), a sensor module (3120), a vision module (3130), a location information extraction module (3140), a location information reset module (3150), a DB update module (3160), and a DB monitoring module (3170), and the cloud server (2200) may include at least one of a DB backup module (3210) and a DB modification module (3220).

The above autonomous driving robot (3100) can operate in a manner that is connected to a cloud server (3200) through an Internet communication network and interoperates with the cloud server (3200).

Below, each component is described in detail in the order of the DB module (3110), sensor module (3120), vision module (3130), location information extraction module (3140), location information reset module (3150), DB update module (3160), and DB monitoring module (3170) configured in the autonomous driving robot (3100), and the DB backup module (3210) and DB modification module (3220) configured in the cloud server (3200).

The above DB module (3110) can store a plurality of pre-generated vision data and a plurality of standard location information by matching them with each other, and can store them as a plurality of location groups each grouped according to a set area having a certain coordinate range.

Such DB module (3110) may include at least one of a location group creation unit (3111) and a location group management unit (3112), as illustrated in FIG. 15.

The above location group generation unit (3111) can generate data sets by matching DB vector data extracted from each vision data (image of a driving video) and standard location information representing coordinate information where each vision data is generated, and can group multiple generated data sets according to a set area having a certain coordinate range to generate multiple location groups.

For example, as illustrated in Fig. 5, each location group can be defined for each setting area 'S1' (or setting area 'S2' wider than 'S1') consisting of a certain coordinate range. In one location group, a plurality of standard location information (location), that is, standard coordinate data, are defined at certain distance intervals, and an image matching each standard coordinate is defined as vector data.

The above location group management unit (3112) can store a data set of a location group created through a location group creation unit (3111), and when a data update command and data to be updated are input by a DB update module (3160), it can modify previously stored data.

The above sensor module (3120) calculates estimated location information (information estimating the location of the robot itself) of the autonomous driving robot (3100) through LiDAR and Odometry systems, and can specify one location group among a number of location groups defined in the DB module (3110) based on the calculated estimated location information.

The above sensor module (3120) can compare the estimated location information of the autonomous driving robot (3100) calculated through LiDAR and Odometry with the set area defined for each location group to determine the location group corresponding to the estimated location information of the autonomous driving robot (3100).

As described above, since each location group has a defined location range, i.e., a coordinate range, the sensor module (3120) can check which location group the estimated location information of the autonomous driving robot (3100) belongs to and output the estimated location information and the location group ID or identification information corresponding to the estimated location information.

The above vision module (3130) can provide a vision recognition result through a vision recognition process for a driving image of an autonomous driving robot (3100).

To this end, the vision module (3130) may include at least one of a driving image generation unit (3131), a driving image acquisition unit (3132), and a vector data extraction unit (3133), as illustrated in FIG. 16.

The above driving image generation unit (3131) can generate driving images of the autonomous driving robot (3100) through at least two cameras installed in the autonomous driving robot (3100) and shooting images in different directions or angles.

For example, to increase the accuracy of image similarity comparison, driving images of an autonomous driving robot (3100) can be generated using a camera facing the ceiling and a camera facing the front, respectively.

The above driving image acquisition unit (3132) can acquire images of driving images of the autonomous driving robot (3100) at regular intervals and assign an image ID to each acquired image.

The above vector data extraction unit (3133) can extract current vector data from each driving image acquired through the driving image acquisition unit (3132) and provide it as a vision recognition result.

The above location information extraction module (3140) can find the vision data (DB vector data) most similar to the vision recognition result (current vector data) among the vision data (DB vector data) belonging to a specific location group through the sensor module (3120), and extract standard location information matching the vision data (DB vector data).

To this end, the location information extraction module (3140) may include at least one of a data input unit (3141), a location group selection unit (3142), a similarity result calculation unit (3143), and a location information provision unit (3144), as illustrated in FIG. 17.

The above data input unit (3141) can receive current vector data from the vision module (3130) and input estimated location information and specific location group information (location group ID or identification information) from the sensor module (3120), respectively. Accordingly, information on the image currently being captured through the vision module (3130), location information recognized through the sensor module (3120), and information on the location group corresponding to the location information can be input, respectively.

The above location group selection unit (3142) can select a location group stored in the DB module (3110) based on location group information (location group ID or identification information) input through the data input unit (3141).

The above similarity result producing unit (3143) can perform a cosine similarity measurement between 'DB vector data' belonging to a location group selected through a location group selecting unit (3142) and 'current vector data' input through a data input unit (3141) and produce a score accordingly.

The above location information providing unit (3144) checks whether the highest similarity score among the similarity scores calculated through the similarity result calculating unit (3143) exceeds a preset threshold score, and if the highest similarity score exceeds the threshold score, extracts and provides standard location information that matches the DB vector data that received the highest similarity score.

The above location information reset module (3150) can compare the standard location information extracted through the location information extraction module (3140) and the estimated location information calculated through the sensor module (3120), and reset the location information of the autonomous driving robot (10) based on the comparison result.

More specifically, the location information reset module (3150) compares the estimated location information produced by the sensor module (3120) with the standard location information provided by the location information provider (3144) to determine whether they match, and if they do not match, determines that there is an abnormality in the driving status of the autonomous driving robot (3100), resets the LiDAR and Odometry of the sensor module (3120) while the autonomous driving robot (3100) is temporarily stopped, and then sets the standard location information provided by the location information provider (3144) as the initial location information of the LiDAR and Odometry to correct the location error.

This location information reset module (3150) performs a similarity check between an image acquired in real time through visual loop detection and a pre-built database, and can readjust the initial location information from LiDAR and Odometry based on the location information estimated using visual information when there is an abnormality in the driving status of the autonomous driving robot (3100).

The above DB update module (3160) adds the image of the current vector data having the highest similarity score to the candidate DB if the highest similarity score does not exceed a threshold, and then performs a similarity comparison with vector data previously added to the candidate DB. If the similarity score exceeds the threshold, it performs an up-count on the image ID of the corresponding vector data, and if the similarity score does not exceed the threshold, it assigns a new image ID and adds it to the candidate DB. However, if the cumulative counter number for the corresponding image ID exceeds a preset threshold number by repeatedly performing the cosine similarity score calculation process and the threshold score comparison process, the DB module (3110) can additionally register the average value of the current vector data having the highest similarity score and the estimated location information corresponding to the current vector data to the DB module (3110) to update the data set of the DB module (3110).

The above DB monitoring module (3170) can perform monitoring on data (data set including DB vector data and standard location information) stored in the DB module (3110) and generate DB monitoring data.

More specifically, the DB monitoring module (3170) can monitor the initial creation time information of vision data, the most recent time information selected by the highest similarity score for each vision data, the number of times information selected by the highest similarity score for each vision data, and the comparison information on the distance and direction difference between the standard location information selected by the highest similarity score and the estimated location information of the sensor module (3120).

The above DB backup module (3210) can be synchronized with the DB module (3110) of the autonomous driving robot (3100) and perform the role of backing up and managing data (data set including DB vector data and standard location information) stored in the DB module (3110).

The above DB modification module (3220) can perform the role of determining data that is not necessary for the DB module (3110) based on DB monitoring data provided from the DB monitoring module (3170) of the autonomous driving robot (3100) and deleting and managing the corresponding data.

In this way, the final data (data set including DB vector data and standard location information) stored in the autonomous driving robot (3100) are stored in the DB backup module (3210) of the cloud server (3200) at regular intervals, and by performing a modification task through an administrator on the data stored in the DB backup module (3210) and then transmitting it to the autonomous driving robot (3100), the modified data can be applied to the autonomous driving robot (3100).

Modification of data in the DB module (3110) can increase the reliability of data stored in the DB module (3110) by removing data that is no longer needed and data with errors according to the changing surrounding environment in addition to the DB automatic update algorithm that is performed locally, i.e., in the autonomous driving robot (3100). At this time, the administrator can modify the data in the DB module (3110) by selecting data to be removed or removing data that satisfies specific conditions through an algorithm, thereby applying it to the autonomous driving robot (3100).

FIG. 18 is a drawing showing the overall configuration and configuration form of a SLAM position error correction system according to the fourth embodiment of the present invention, FIG. 19 is a block diagram showing the configuration of a first DB module according to the fourth embodiment of the present invention, FIG. 20 is a block diagram showing the configuration of a vision module according to the fourth embodiment of the present invention, and FIG. 31 is a block diagram showing the configuration of a first position information extraction module according to the fourth embodiment of the present invention.

Referring to FIG. 18, the SLAM position error correction system (4000) according to the fourth embodiment may include at least one of an autonomous driving robot (4100) and a cloud server (4200).

The above autonomous driving robot (4100) may include at least one of a first DB module (4110), a sensor module (4120), a vision module (4130), a first location information extraction module (4140), a location information reset module (4150), and a DB update module (4160), and the cloud server (2200) may include at least one of a second first DB module (4210) and a second location information extraction module (4220).

The above autonomous driving robot (4100) can operate in a manner that is connected to a cloud server (4200) through an Internet communication network and interoperates with the cloud server (4200).

Below, each component is described in detail in the order of the first DB module (4110), sensor module (4120), vision module (4130), first location information extraction module (4140), location information reset module (4150), and DB update module (4160) configured in the autonomous driving robot (4100), and the second DB module (4210) and second location information extraction module (4220) configured in the cloud server (4200).

The above first DB module (4110) can store a plurality of pre-generated vision data and a plurality of standard location information by matching them with each other, and can store them as a plurality of location groups each grouped according to a set area having a certain coordinate range.

This first DB module (4110) may include at least one of a location group creation unit (4111) and a location group management unit (4112), as illustrated in FIG. 19.

The above location group generation unit (4111) matches the DB vector data extracted from each vision data (image of a driving video) and the standard location information representing the coordinate information where each vision data is generated to generate each data set, and groups the generated multiple data sets according to a set area having a certain coordinate range to generate multiple location groups. Here, the set area means a range for selecting a location group for performing a similarity search at the location of the autonomous driving robot (4100), and standard location information that is divided and fixed at a certain interval is defined in one location group.

For example, as illustrated in FIG. 5, in the case of a setting area 'S1' (corresponding to one location group) consisting of a certain coordinate range, since the location where the autonomous driving robot (4100) is displayed is in the center of the group, a similarity comparison search only needs to be performed in that group, and in the case of 'S2', since the location where the autonomous driving robot (4100) is displayed is located at the boundary of the group, it means that the surrounding groups need to be searched as well, and accordingly, a group where a similarity comparison search is to be performed can be selected relatively more widely, such as 'S2'.

The above location group management unit (4112) can store a data set of a location group created through a location group creation unit (4111), and when a data update command and data to be updated are input by a DB update module (4160), it can modify previously stored data.

The above sensor module (4120) calculates estimated location information (information estimating the location of the robot itself) of the autonomous driving robot (4100) through a LiDAR and Odometry system, and can specify one location group among a plurality of location groups defined in the first DB module (4110) based on the calculated estimated location information.

The above sensor module (4120) can compare the estimated location information of the autonomous driving robot (4100) calculated through LiDAR and Odometry with the set area defined for each location group to determine the location group corresponding to the estimated location information of the autonomous driving robot (4100).

As described above, since each location group has a defined location range, i.e., a coordinate range, the sensor module (4120) can check which location group the estimated location information of the autonomous driving robot (4100) belongs to and output the estimated location information and the location group ID or identification information corresponding to the estimated location information.

The above vision module (4130) can provide a vision recognition result through a vision recognition process for a driving image of an autonomous driving robot (4100).

To this end, the vision module (4130) may include at least one of a driving image generation unit (4131), a driving image acquisition unit (4132), and a vector data extraction unit (4133), as illustrated in FIG. 20.

The above driving image generation unit (4131) can generate driving images of the autonomous driving robot (4100) through at least two cameras installed in the autonomous driving robot (4100) and shooting images in different directions or angles.

For example, to increase the accuracy of image similarity comparison, driving images of an autonomous driving robot (4100) can be generated using a camera facing the ceiling and a camera facing the front, respectively.

The above driving image acquisition unit (4132) can acquire images of driving images of the autonomous driving robot (4100) at regular intervals and assign an image ID to each acquired image.

The above vector data extraction unit (4133) can extract current vector data from each driving image acquired through the driving image acquisition unit (4132) and provide it as a vision recognition result.

The above first location information extraction module (4140) can find the vision data (DB vector data) most similar to the vision recognition result (current vector data) among the vision data (DB vector data) belonging to a specified location group through the sensor module (4120), and extract standard location information matching the corresponding vision data (DB vector data).

To this end, the first location information extraction module (4140) may include at least one of a data input unit (4141), a location group selection unit (4142), a similarity result calculation unit (4143), and a location information provision unit (4144), as illustrated in FIG. 21.

The above data input unit (4141) can receive current vector data from the vision module (4130) and input estimated location information and specific location group information (location group ID or identification information) from the sensor module (4120), respectively. Accordingly, information on the image currently being captured through the vision module (4130), location information recognized through the sensor module (4120), and information on the location group corresponding to the location information can be input, respectively.

The above location group selection unit (4142) can select a location group stored in the first DB module (4110) according to the location group information (location group ID or identification information) input through the data input unit (4141). In this way, instead of performing a comparison process on the entire DB image to find an image similar to the image recognized through the vision module (4130), by specifying a group belonging to the location recognized through the sensor module (4120), candidate images to be compared with the image recognized by the vision module (4130) can be specified, thereby reducing the number of comparison targets. Accordingly, since the number of comparisons is reduced, the time required for the similarity comparison process can be reduced, and errors such as cases where images are similar but located in different places can be eliminated, thereby increasing accuracy.

The above similarity result producing unit (4143) can perform a cosine similarity measurement between the 'DB vector data' belonging to the location group selected through the location group selecting unit (4142) and the 'current vector data' input through the data input unit (4141) and produce a score accordingly.

The above location information providing unit (4144) checks whether the highest similarity score among the similarity scores calculated through the similarity result calculating unit (4143) exceeds a preset threshold score, and if the highest similarity score exceeds the threshold score, extracts and provides standard location information that matches the DB vector data that received the highest similarity score.

The above location information reset module (4150) can compare the standard location information extracted through the first location information extraction module (4140) and the estimated location information calculated through the sensor module (4120), and reset the location information of the autonomous driving robot (4100) based on the comparison result.

More specifically, the location information reset module (4150) compares the estimated location information produced by the sensor module (4120) with the standard location information provided by the location information provider (4144) to determine whether they match, and if they do not match, determines that there is an abnormality in the driving status of the autonomous driving robot (4100), resets the LiDAR and Odometry of the sensor module (3000) while the autonomous driving robot (4100) is temporarily stopped, and then sets the standard location information provided by the location information provider (4144) as the initial location information of the LiDAR and Odometry to correct the location error.

This location information reset module (4150) performs a similarity check between an image acquired in real time through visual loop detection and a pre-built database, and can readjust the initial location information from LiDAR and Odometry based on the location information estimated using visual information when there is an abnormality in the driving status of the autonomous driving robot (4100).

The above DB update module (4160) adds the image of the current vector data having the highest similarity score to the candidate DB if the highest similarity score does not exceed a threshold, and then performs a similarity comparison with vector data previously added to the candidate DB. If the image exceeds the threshold, an up-count is performed on the image ID of the corresponding vector data. If the image does not exceed the threshold, a new image ID is assigned and added to the candidate DB. However, if the cumulative counter count for the corresponding image ID exceeds a preset threshold count by repeatedly performing the cosine similarity score calculation process and the threshold score comparison process, the average value of the current vector data having the highest similarity score and the estimated location information corresponding to the current vector data is additionally registered in the first DB module (4110), so as to update the data set of the first DB module (4110).

The above second DB module (4210) is a component that stores a plurality of pre-generated vision data (DB vector data) and a plurality of standard location information by matching them with each other, and stores them as a plurality of location groups each grouped according to a set area having a certain coordinate range, and has more data (data set including DB vector data and standard location information) than the first DB module (4110) installed in the autonomous driving robot (4100), so that when the autonomous driving robot (4100) is unable to perform the first location estimation, the cloud server (4200) can perform the second location estimation for the autonomous driving robot (4100). This second DB module (4210) is constructed with the same data structure as the above-described first DB module (4210), but by utilizing the advantages of the cloud, more data is constructed so as to provide high accuracy for location estimation.

The second location information extraction module (4220) above can extract standard location information that matches the vision data most similar to the vision recognition result among the vision data belonging to a specific location group through the sensor module (4120) of the autonomous driving robot (4100) when the standard location information is not extracted through the first location information extraction module (4140) of the autonomous driving robot (4100) (i.e., when the location estimation is not possible in the autonomous driving robot (4100), i.e., perform a second location estimation process, and extract the standard location information extracted accordingly and provide it to the first location information extraction module (4140) of the autonomous driving robot (4100). However, when location estimation (standard location information extraction) is not possible even in the cloud server (4200), the DB can be updated through the DB automatic update algorithm.

These cloud servers (4200) continuously receive and accumulate data on driving records, judgment results, etc. of a plurality of autonomous driving robots (4100) and form a composite DB, thereby helping to determine the driving status of the autonomous driving robot (4100) through virtual driving and recommending a driving route while the robot is driving.

Claims (17)

A DB module that stores a plurality of pre-generated vision data and a plurality of standard location information by matching them with each other, and stores them as a plurality of location groups, each grouped according to a set area having a certain coordinate range; A sensor module that calculates estimated location information of an autonomous driving robot using LiDAR and Odometry and specifies one location group among multiple location groups based on the calculated estimated location information; A vision module that provides vision recognition results through a vision recognition process for driving images of autonomous driving robots; A location information extraction module that extracts standard location information that matches the vision data most similar to the vision recognition result among the vision data belonging to a specific location group through the sensor module; and A device capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM (Simultaneous Localization And Map-Building) technology, characterized in that it includes a position information reset module that compares the standard position information extracted through the above position information extraction module and the estimated position information produced through the above sensor module, and resets and corrects the position information of the autonomous driving robot according to the comparison result. In the first paragraph, The above DB module, A location group generation unit that creates a data set by matching the DB vector data extracted from each vision data and the standard location information representing the coordinate information where each vision data is generated, and groups the generated multiple data sets according to the setting area to create multiple location groups; and A device capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology, characterized by including a position group management unit that stores a data set of a position group generated through the above position group generation unit. In the second paragraph, The above vision module, A driving image generation unit that generates driving images of the autonomous driving robot through at least two cameras installed on the autonomous driving robot and shooting from different directions or angles; A driving image acquisition unit that acquires images of driving images of an autonomous driving robot and assigns an image ID to each acquired image; and A device capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology, characterized by including a vector data extraction unit that extracts current vector data from each image acquired through the driving image acquisition unit and provides it as a vision recognition result. In the third paragraph, The above location information extraction module, A data input unit that receives current vector data from the vision module and inputs estimated location information and specific location group information from the sensor module; A location group selection unit that selects a location group stored in the DB module based on location group information entered through the data input unit; A similarity result calculation unit that calculates a cosine similarity score between the DB vector data belonging to the location group selected through the above location group selection unit and the current vector data input through the above data input unit; and A device capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology, characterized in that it includes a position information providing unit that checks whether the highest similarity score among the similarity scores calculated through the similarity result calculating unit exceeds a preset reference score, and if so, extracts and provides standard position information matched with the DB vector data of the highest similarity score. In the fourth paragraph, The above location information reset module is, A device capable of correcting position errors through real-time analysis and comparison between vision data and LiDAR data for implementing SLAM technology, characterized in that the estimated position information produced through the sensor module is compared with the standard position information provided through the position information provision unit to determine whether they match, and if they do not match, it is determined that there is an abnormality in the driving status of the autonomous driving robot, so that the autonomous driving robot is temporarily stopped, the LiDAR and odometry of the sensor module are reset, and then the standard position information provided through the position information provision unit is set as the initial position information of the LiDAR and odometry to correct the position error. In the fourth paragraph, The above device, A device capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology, characterized in that the device further includes a DB update module for adding the image of the current vector data having the highest similarity score to the candidate DB if the highest similarity score does not exceed the reference score, and then performing a similarity comparison with vector data previously added to the candidate DB, and if it exceeds the reference score, performing an up-count for the image ID of the corresponding vector data, and if it does not exceed the reference score, assigning a new image ID and adding it to the candidate DB, but repeatedly performing the cosine similarity score calculation process and the reference score comparison process, and when the accumulated counter count for the corresponding image ID exceeds a preset reference count, additionally registering an average value of the current vector data having the highest similarity score and the estimated position information corresponding to the current vector data to the DB module to update the data set of the DB module. In the first paragraph, The above sensor module, A device capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology, characterized in that the estimated position information of an autonomous driving robot calculated through lidar and odometry is compared with the above-described setting area defined by position group to determine a position group corresponding to the estimated position information of the autonomous driving robot. An autonomous driving robot including a sensor module and a vision module; and It includes a DB module, a location information extraction module and a location information reset module, and includes a cloud server connected to the autonomous driving robot through an Internet communication network. The above DB module stores a plurality of pre-generated vision data and a plurality of standard location information by matching them with each other, and stores them as a plurality of location groups each grouped according to a set area having a certain coordinate range. The above sensor module calculates estimated location information of the autonomous driving robot through LiDAR and Odometry, and specifies one location group among multiple location groups based on the calculated estimated location information. The above vision module provides vision recognition results through a vision recognition process for the driving video of the autonomous driving robot. The above location information extraction module extracts standard location information that matches the vision data most similar to the vision recognition result among the vision data belonging to the specified location group through the sensor module, The above location information reset module is characterized in that it compares the standard location information extracted through the location information extraction module and the estimated location information produced through the sensor module, and resets and corrects the location information of the autonomous driving robot according to the comparison result, and is a system capable of correcting location errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM (Simultaneous Localization And Map-Building) technology. In Article 8, The above DB module, A location group generation unit that creates a data set by matching the DB vector data extracted from each vision data and the standard location information representing the coordinate information where each vision data is generated, and groups the generated multiple data sets according to the setting area to create multiple location groups; and A system capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology, characterized by including a position group management unit that stores a data set of a position group generated through the above position group generation unit. In Article 9, The above vision module, A driving image generation unit that generates driving images of the autonomous driving robot through at least two cameras installed on the autonomous driving robot and shooting from different directions or angles; A driving image acquisition unit that acquires images of driving images of an autonomous driving robot and assigns an image ID to each acquired image; and A system capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology, characterized in that it includes a vector data extraction unit that extracts current vector data from each image acquired through the driving image acquisition unit and provides it as a vision recognition result. In Article 10, The above location information extraction module, A data input unit that receives current vector data from the vision module and inputs estimated location information and specific location group information from the sensor module; A location group selection unit that selects a location group stored in the DB module based on location group information entered through the data input unit; A similarity result calculation unit that calculates a cosine similarity score between the DB vector data belonging to the location group selected through the above location group selection unit and the current vector data input through the above data input unit; and A system capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology, characterized in that it includes a position information providing unit that checks whether the highest similarity score among the similarity scores calculated through the similarity result calculating unit exceeds a preset reference score, and if so, extracts and provides standard position information matched with the DB vector data of the highest similarity score. In Article 11, The above location information reset module is, A system capable of correcting position errors through real-time analysis and comparison between vision data and LiDAR data for implementing SLAM technology, characterized in that the estimated position information produced through the sensor module is compared with the standard position information provided through the position information provision unit to determine whether they match, and if they do not match, it is determined that there is an abnormality in the driving status of the autonomous driving robot, so that the autonomous driving robot is temporarily stopped, the LiDAR and odometry of the sensor module are reset, and then the standard position information provided through the position information provision unit is set as the initial position information of the LiDAR and odometry to correct the position error. In Article 11, The above cloud server, A system capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology, characterized in that the system further includes a DB update module for adding the image of the current vector data having the highest similarity score to the candidate DB if the highest similarity score does not exceed the reference score, and then performing a similarity comparison with vector data previously added to the candidate DB, and if it exceeds the reference score, performing an up-count on the image ID of the corresponding vector data, and if it does not exceed the reference score, assigning a new image ID and adding it to the candidate DB, but repeatedly performing the cosine similarity score calculation process and the reference score comparison process, and if the accumulated counter count for the corresponding image ID exceeds a preset reference count, additionally registering an average value of the current vector data having the highest similarity score and the estimated position information corresponding to the current vector data in the DB module to update the data set of the DB module. In Article 8, The above sensor module, A system capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology, characterized in that the estimated position information of an autonomous driving robot calculated through lidar and odometry is compared with the above-described setting area defined by position group to determine a position group corresponding to the estimated position information of the autonomous driving robot. An autonomous driving robot comprising: a DB module that stores a plurality of pre-generated vision data and a plurality of standard location information by matching them with each other, and stores them as a plurality of location groups each grouped according to a set area having a certain coordinate range; a sensor module that calculates estimated location information of an autonomous driving robot through LiDAR and Odometry and specifies one of the plurality of location groups according to the calculated estimated location information; a vision module that provides a vision recognition result through a vision recognition process for a driving image of the autonomous driving robot; a location information extraction module that extracts standard location information that matches the vision data most similar to the vision recognition result among the vision data belonging to the location group specified through the sensor module; a location information reset module that compares the standard location information extracted through the location information extraction module and the estimated location information calculated through the sensor module, and resets and corrects the location information of the autonomous driving robot according to the comparison result; and a DB monitoring module that performs monitoring on data stored in the DB module and generates DB monitoring data; and A system capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM (Simultaneous Localization And Map-Building) technology, characterized by including a DB backup module that is synchronized with the DB module and backs up data stored in the DB module; and a DB modification module that determines and deletes data that is not necessary for the DB module based on DB monitoring data provided from the DB monitoring module, and including a cloud server connected to the autonomous driving robot through an Internet communication network. In Article 15, The above DB monitoring module, A system capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementing SLAM technology, characterized in that it monitors the initial generation time information of vision data, the most recent time information selected by the highest similarity score for each vision data, the number of times information selected by the highest similarity score for each vision data, and the comparison information on the distance and direction difference between the standard location information selected by the highest similarity score and the estimated location information of the sensor module, respectively. An autonomous driving robot comprising: a first DB module that matches and stores a plurality of pre-generated vision data and a plurality of standard location information with each other, and stores them as a plurality of location groups each grouped according to a set area having a certain coordinate range; a sensor module that calculates estimated location information of the autonomous driving robot through LiDAR and Odometry and specifies one of the plurality of location groups according to the calculated estimated location information; a vision module that provides a vision recognition result through a vision recognition process for a driving image of the autonomous driving robot; a first location information extraction module that extracts standard location information that matches the vision data most similar to the vision recognition result among the vision data belonging to the location group specified through the sensor module; and a location information reset module that compares the standard location information extracted through the first location information extraction module and the estimated location information calculated through the sensor module, and resets and corrects the location information of the autonomous driving robot according to the comparison result; and A system capable of correcting position errors through real-time analysis and comparison between vision data and lidar data for implementation of SLAM (Simultaneous Localization And Map-Building) technology, characterized in that it includes a second DB module that matches and stores a plurality of pre-generated vision data and a plurality of standard location information, and stores them as a plurality of location groups each grouped according to a set area having a certain coordinate range; and a second location information extraction module that extracts standard location information that matches the vision data most similar to the vision recognition result among the vision data belonging to the specified location group through the sensor module when the standard location information is not extracted through the first location information extraction module and provides the extracted standard location information to the first location information extraction module, and includes a cloud server connected to the autonomous driving robot through an Internet communication network.

Images