WO2006059419A1 - 追跡装置および追跡方法 - Google Patents
追跡装置および追跡方法 Download PDFInfo
- Publication number
- WO2006059419A1 WO2006059419A1 PCT/JP2005/016711 JP2005016711W WO2006059419A1 WO 2006059419 A1 WO2006059419 A1 WO 2006059419A1 JP 2005016711 W JP2005016711 W JP 2005016711W WO 2006059419 A1 WO2006059419 A1 WO 2006059419A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- target
- image data
- tracking
- feature data
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/62—Extraction of image or video features relating to a temporal dimension, e.g. time-based feature extraction; Pattern tracking
Definitions
- the present invention relates to a tracking device and a tracking method for capturing a moving image and tracking an object.
- Coexistence robots need to recognize the environment and human actions, and if they move and follow things with their eyes, human visual functions must be realized as a system.
- systems that can quickly and accurately obtain information on surveillance cameras and video surveillance that examines recorded video, as well as human behavior. increasing.
- the inter-frame difference image force also extracts the area of the moving object by extracting the edge part of the moving object to be tracked, and the direction of the point or figure in the image at the next moment.
- an optical flow which is a velocity vector indicating how much distance to move, and selects only the pixels with a large flow as the region of the moving object.
- Patent Document 1 discloses an object tracking method that uses both motion information based on correlation (matching) between an optical flow and a template image.
- Patent Document 1 Japanese Patent Application Laid-Open No. 2004-240762
- the object of the present invention is to solve such problems and realize real-time tracking with a method that is as light as possible and recognizes it by sequentially learning and updating target features through tracking. It provides a tracking device and a tracking method that realizes more accurate tracking at the level.
- tracking is performed while identifying an object by using an object-specific feature that does not match image (pixel) level matching (correlation) found in many conventional methods.
- object-specific feature that does not match image (pixel) level matching (correlation) found in many conventional methods.
- correlation image level matching
- the tracking device of the present invention divides input moving image data to generate divided image data, and extracts non-background image data of a portion different from background image data from the divided image data.
- Extracting means for determining, based on the output of the extracting means, target presence / absence determining means for determining whether the divided image includes at least a part of the tracking target, non-background image determined to have the target
- feature data calculation means for calculating color higher-order local autocorrelation feature data
- target position determination means for determining the position of a tracking target by grouping adjacent divided images for divided images determined to have a target
- an adder / synthesizer for adding and synthesizing the feature data of the collected non-background image data, and a distance between the registered feature data of the tracking target and the output of the adder / synthesizer.
- key comprising the identifying means for identifying Dzu-out tracked.
- the tracking device of the present invention includes an image dividing unit that divides input moving image data to generate divided image data, and a non-background image of a portion different from background image data from the divided image data.
- Extraction means for extracting data
- feature data calculation means for calculating color higher-order local autocorrelation feature data for non-background image data
- color higher-order local autocorrelation feature data force norm which is the length of feature data vector Based on the output of the norm calculation means
- a target presence / absence determination means for determining whether at least a part of the tracking target is included in the divided image, and an amount determined to be present.
- target position determining means for determining the target position by collecting adjacent divided images
- addition composition means for adding and combining the feature data of the collected non-background image data
- registration The main feature is that it comprises an identification means for identifying the tracking object based on the distance between the feature data of the tracking object being recorded and the output of the adding and synthesizing means.
- the tracking device described above may include background image update means for updating a background image based on the divided image determined by the target presence / absence determination means that there is no tracking target.
- the identification unit includes a storage unit that stores a plurality of latest feature data for each tracking target, all stored feature data, and feature data of the detected target. Each distance is calculated, the distance is the nearest !, the latest feature data extraction means for extracting the odd number of feature data, and the detected object is the number of feature data in the extracted odd number of feature data.
- an object identification means for determining that the object belongs to the tracking target with the largest number.
- registration is performed based on the identification result of the identification means! It also has a registration target update means for updating the feature data of the tracking target.
- Higher-order local autocorrelation features are known to be effective features for personal identification such as face recognition in conventional research.
- this higher-order local autocorrelation feature is further extended to color, and a color higher-order local autocorrelation feature that extracts shape and color information simultaneously is used.
- This color higher-order local autocorrelation feature has position invariance in the image, and can be accurately recognized no matter where the tracking target is in the divided image. Therefore, tracking is performed while identifying the tracking target at the feature level without the need to accurately detect or predict the target itself (contour) as in conventional image matching, thereby realizing more robust and highly accurate tracking. There is an effect that can be done.
- the method of the present invention is equivalent to performing the process of tracking and the process of acquiring information from the tracking target at the same time. Therefore, even if the target cannot be recognized due to hiding or crossing with another target, tracking can be continued without any problems by identifying the target after the status is resolved. Another advantage is that real-time processing with a small amount of calculation for feature extraction is possible. Further, since each divided image can be processed in parallel, the processing speed can be further improved by performing parallel processing using a PC cluster or the like.
- the method of the present invention does not compare with the template image, it is not necessary to prepare a template in advance, thereby improving versatility and accurate feature data of the newly detected object. Can be acquired immediately, and does not depend on the size or movement of the target image, and can be traced with high accuracy even if the shape, color, and size change as well as the position over time. There is also.
- FIG. 1 is a block diagram showing a configuration of a tracking device according to the present invention.
- FIG. 2 is a flowchart showing the contents of the object tracking process of the present invention.
- FIG. 3 is an explanatory diagram showing types of displacement vectors of color higher-order local autocorrelation.
- FIG. 4 is an explanatory diagram showing an example of an input image divided into strips.
- FIG. 5 is an explanatory view showing a non-background image example (whole).
- FIG. 6 is a flowchart showing the contents of the tracking process of the second embodiment of the present invention. Explanation of symbols
- the following processing is performed to estimate the position of the tracking target.
- the image is divided into strips (or rectangles), and by checking each strip-shaped divided image, a divided image in which the tracking target exists is extracted.
- a divided image in which the tracking target exists is extracted by comparing a background image not including the tracking target with the current image.
- the current position of the tracking target can be acquired by grouping adjacent images that are the tracking target.
- feature data for the tracking target is generated and compared with the registered data to identify the tracking target. Acquire features while tracking the tracking target, identify the target using the features, and realize high-precision tracking.
- FIG. 1 is a block diagram showing a configuration of a tracking device according to the present invention.
- the video camera 10 outputs color moving image frame data of the target person or device in real time.
- the color moving image data may be 256 gradations of RGB.
- a computer 12 such as a PC (computer) captures color moving image frame data from the video camera 10 in real time via an external or built-in video capture device 11 for capturing moving images.
- the computer 12 is connected to a known monitor device 13, keyboard 14, mouse 15, and LAN 20.
- the LAN 30 is connected to the Internet 30 via the router 23, and the computer 12 can communicate with other PCs 21 and 22 on the LAN and the PC 31 on the Internet.
- the tracking device of the present invention is realized by creating, installing, and starting a program for executing processing described later on a known arbitrary computer 11 such as a personal computer.
- the other PCs 21, 22, and 31 can be used as a parallel processing device in the tracking process or for remote display of the tracking result.
- the video camera 10 may input the moving image data input in real time to disclose an example in which the moving image data is stored in a moving image file and may be read and processed sequentially. .
- FIG. 2 is a flowchart showing the contents of the object tracking process of the present invention executed in the computer 12.
- an example of tracking a person as a tracking target is opened.
- the present invention is applicable to any tracking object.
- image frame data is captured in real time from the video camera 10 using the video capture device 11.
- the input image data is divided into N strip-shaped (divided only vertically) or rectangular (divided horizontally and vertically) images. At the dividing boundary, it is cut out redundantly by one pixel more than the boundary line in order to calculate the feature data described later.
- the size of the divided image should be about the same size as the target and smaller. When there are a large number of objects, the separation accuracy of the objects is further improved by reducing the size of the divided image.
- FIG. 4 is an explanatory diagram showing an example of an image that is input to the computer 12 and divided into strips.
- the tracking target extraction method include a method using an optical flow, a method using an interframe difference method, and a method using a background difference method.
- the method using optical flow is not very suitable for extraction of the tracking target because it is vulnerable to noise with a large amount of calculation.
- the interframe difference method has a problem that the target cannot be extracted if the target is stationary.
- the background subtraction method can extract accurately even if the tracking target is stationary. Therefore, the background subtraction method is used in the present invention.
- the background subtraction method is a method for obtaining an object to be processed by taking a difference between a background image prepared in advance by some method and the current input image.
- a background image an image that does not include a tracking target may be input and stored in advance.
- tracking targets such as roads and places with many traffics, and it is difficult to obtain images that do not include tracking targets. Even in such a case, it is necessary to estimate the background image without including the tracking target. Therefore, multiple background images that do not use the background image as it is You may take the median on a time series about each pixel of an image.
- the background image is updated by replacing an image having no tracking target in strip units with a background image.
- FIG. 5 is an explanatory diagram showing an example of a non-background image (entire) extracted from a plurality of divided images. Extracted even though there is no target to be traced, as indicated by the circled circle in the figure! There is “noise”. In order to reduce this noise as much as possible, the following processing is performed.
- the background image is B (x)
- the threshold ⁇ is a small value
- the difference image G (X) is as follows: To calculate.
- This process is performed for RGB, and a difference image G is generated. This process does not detect slight changes, reduces noise, and improves the accuracy of target detection. Note that the background image changes over time due to the effects of lighting, sunlight, and shadows. Such a change cannot be dealt with only by the processing described above, so the background image is always updated by the processing described later.
- S14 a non-background area is obtained. That is, since the difference pixel value is 0 in the background portion, the number of pixels other than 0 in the strip-like difference image is counted.
- the counting process in S14 may be performed simultaneously with the determination process for each pixel in S13.
- S15 it is determined whether there is a high possibility that the target exists depending on whether the value of the non-background area (number of pixels) for each strip is larger than a predetermined threshold! If the result is negative, move to S17. If positive, move to S16.
- the optimum value of the predetermined threshold value may be determined by experiment, but it must be detected even when the target is divided from the adjacent strip, so 1Z20 of the total number of pixels of the tracking target (person) image. About 1Z5.
- color higher-order local autocorrelation features are calculated and stored.
- high-order local autocorrelation which is a feature that has been proven in personal identification, can extract shape information.
- the feature has been extended to color so that both color and shape information can be extracted simultaneously.
- the higher-order autocorrelation feature is an extension of the autocorrelation feature to a higher order. If the target image in the screen is f (r), the higher-order local autocorrelation feature is the displacement direction (a,-, a )
- High-order local autocorrelation features are considered innumerable depending on the order and the direction of displacement (a,-, a).
- the displacement vector a takes values such as (0,0), (0,1), (1,0), (1, 1),.
- the feature data calculated in this way has the following properties.
- the feature value of the entire image is the sum of the feature values of each target in the image. In other words, the sum of the features of each part of the target is equal to the features of the entire target.
- S17 it is determined whether or not there is noise and whether or not there is noise depending on whether or not the value of the non-background area (number of pixels) for each strip is larger than a predetermined second threshold value! If the judgment result is negative, then this is the force to move to S18.
- the background data is updated using background data in which there is no target and there is almost no noise. As an update method, it may be simply overwritten and replaced, or the median of the pixel values of the most recent background data may be adopted.
- S19 it is determined whether or not all the divided image data have been processed. If the determination result is negative, the process proceeds to S12, but if the determination is affirmative, the process proceeds to S20.
- S20 it is determined whether or not the divided image includes at least a part of the target, and for the divided image determined to have the target, the adjacent divided images are grouped to determine the position of the target. At the same time, the feature data of the adjacent divided images are added together to synthesize the target feature data.
- the strip image determined to have a target is further divided into two, “large” and “medium”, according to the size of the non-background area, and the non-adjacent strip images are not covered.
- the size of the background area (“large”, “medium”, “small” (no object))
- a tracking target for example, a bird
- the optimum threshold values for “Large” and “Medium” depend on the tracking target, the size of the input image, the segmentation method, etc., and can be determined by experiment.
- the feature data of the entire object can be obtained simply by adding the feature data that have already been extracted to the strip image in which the subject exists. This is because the feature data is additive.
- the distance to the already registered target feature data is calculated to identify the target.
- normalize the obtained feature data vector X.
- the normality ⁇ here means that the norm is processed so that the norm is 1, specifically, the value of each element of the feature vector X is divided by the norm value.
- the norm (norm) is the length of the feature vector defined by Equation 5 below.
- a point obtained by plotting the extracted target feature on the feature space is placed on a unit sphere centered on the origin in the feature space.
- the distance is 0 and the distance from the origin is 1.
- the closer to the distance force ⁇ the closer the similarity is, and the easier it is to identify. It also provides robust identification against large and small variations due to the distance of the object.
- the k-NN method employed for object identification in the embodiment is the closest to the feature vector to be discriminated while keeping several feature data as they are for all recently detected objects.
- this method k pieces of feature data are extracted, and it is determined that the largest number of pieces of feature data belong to the extracted object. This makes it possible to identify robustly against changes in the shape of the object due to movements such as walking.
- the k-NN method registers a plurality of feature data for each object as they are and compares the distances to each other. Therefore, when the tracking target is a person, a plurality of features with different postures during walking are used. Data will be registered, and gait can be used for identification.
- the target class is determined as A. If the smallest distance is larger than the predetermined value, it is determined that the object is not included in the currently registered error and is registered as a new object. This method keeps multiple pieces of the latest information to be tracked, so it can be tracked stably. Note that registration information increases each time a new target is added, so it is recommended that you remove the powerful registration information that does not detect the target belonging to the registration information even after a certain period of time.
- the tracking result is output.
- the result is information such as the position of the object on the current image and the locus of the object.
- the registered feature data is updated.
- the target class is determined, the oldest registered data of the class is deleted, the latest feature data is registered, and the feature data is updated.
- S24 for example, whether or not to end the process is determined based on whether or not the administrator has performed an end operation. If the determination result is negative, the process proceeds to S10, but if the determination is positive, the process ends. To do.
- Example 2 Example 2
- the presence / absence of the target is determined based on the non-background area of the divided image.
- the background differential image force feature data of all the divided images is calculated, and the feature data Based on the above, the presence or absence of the target is determined.
- FIG. 6 is a flowchart showing the contents of the object tracking process of the second embodiment of the present invention.
- S36 it is determined whether or not there is a possibility that at least a part of the target exists in the divided image depending on whether or not the norm is equal to or greater than a predetermined value, and if the determination result is negative. Will move to S37 if positive. In S37, S18 Similarly, the background data is updated using background data that does not exist and has almost no noise.
- the norm is used to estimate the position of the object.
- the feature norm obtained from the non-background image generated by performing background subtraction is large for images that have a tracking target and is small for images that are not. That is, by comparing the size of the norm of each strip image, it is possible to know which strip image has the tracking target.
- the norm is divided into three types, those whose norm is close to 0, medium, and large.
- a large one is a strip image with a tracking target
- a one near 0 is a tracking image with no tracking target V
- a strip image
- a medium one is a large strip noise or a strip image with (a part of) a tracking target.
- a strip image adjacent to a strip image with a large norm that has a medium norm is considered to be part of the target, and is separated from a strip with a large norm by V, a medium norm.
- a strip with is considered noise.
- the tracking target is a person whose width changes with time, such as a person, it can accurately follow.
- it is possible to acquire the features of the entire target by simply adding the feature data already extracted to the strip image where the tracking target exists. it can.
- the present invention may be modified as follows.
- the detected position information of the target is It is not used for tracking processing. Therefore, it is possible to memorize where the tracking target is now and where it is likely to move next, and add a target position estimation function based on action prediction that is effectively used for later identification. . This can further reduce the computational complexity.
- color and shape information is extracted in the form of color higher-order local autocorrelation features, and the time-based features such as gait are used in the k-NN identification method and the “weight” is used. You may make it.
- complex identification such as color, shape, and movement (gait) is used at a predetermined ratio (weight) for comprehensive identification. By changing (multiplying) the weight according to the importance level, it is possible to identify with higher reliability.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
Description
Claims
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| EP05782365.0A EP1835462A4 (en) | 2004-12-02 | 2005-09-12 | TRACKING DEVICE AND PERSECUTION PROCEDURE |
| US11/792,084 US7957557B2 (en) | 2004-12-02 | 2005-12-09 | Tracking apparatus and tracking method |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2004349244A JP3970877B2 (ja) | 2004-12-02 | 2004-12-02 | 追跡装置および追跡方法 |
| JP2004-349244 | 2004-12-02 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2006059419A1 true WO2006059419A1 (ja) | 2006-06-08 |
Family
ID=36564859
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2005/016711 Ceased WO2006059419A1 (ja) | 2004-12-02 | 2005-09-12 | 追跡装置および追跡方法 |
Country Status (4)
| Country | Link |
|---|---|
| US (1) | US7957557B2 (ja) |
| EP (1) | EP1835462A4 (ja) |
| JP (1) | JP3970877B2 (ja) |
| WO (1) | WO2006059419A1 (ja) |
Families Citing this family (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4368767B2 (ja) * | 2004-09-08 | 2009-11-18 | 独立行政法人産業技術総合研究所 | 異常動作検出装置および異常動作検出方法 |
| JP4215781B2 (ja) * | 2006-06-16 | 2009-01-28 | 独立行政法人産業技術総合研究所 | 異常動作検出装置および異常動作検出方法 |
| JP4603512B2 (ja) * | 2006-06-16 | 2010-12-22 | 独立行政法人産業技術総合研究所 | 異常領域検出装置および異常領域検出方法 |
| JP4429298B2 (ja) * | 2006-08-17 | 2010-03-10 | 独立行政法人産業技術総合研究所 | 対象個数検出装置および対象個数検出方法 |
| US8593506B2 (en) * | 2007-03-15 | 2013-11-26 | Yissum Research Development Company Of The Hebrew University Of Jerusalem | Method and system for forming a panoramic image of a scene having minimal aspect distortion |
| JP4389956B2 (ja) | 2007-04-04 | 2009-12-24 | ソニー株式会社 | 顔認識装置及び顔認識方法、並びにコンピュータ・プログラム |
| JP4842197B2 (ja) * | 2007-04-17 | 2011-12-21 | 財団法人ソフトピアジャパン | 多重分割画像を用いた異常動作検出装置、異常動作検出方法及び異常動作検出用プログラム |
| CN101453660B (zh) * | 2007-12-07 | 2011-06-08 | 华为技术有限公司 | 一种视频目标跟踪方法和装置 |
| JP2010061588A (ja) * | 2008-09-05 | 2010-03-18 | Univ Of Tokyo | 特徴ベクトル算出装置、特徴ベクトル算出方法及びプログラム |
| CN101567087B (zh) * | 2009-05-25 | 2012-05-23 | 北京航空航天大学 | 复杂天空背景下红外序列图像弱小目标检测与跟踪方法 |
| US8340435B2 (en) * | 2009-06-11 | 2012-12-25 | California Institute Of Technology | Method and system for object recognition search |
| JP2012133759A (ja) * | 2010-11-29 | 2012-07-12 | Canon Inc | 侵入物体の検知を行うことができる物体追尾装置、物体追尾方法及び記憶媒体 |
| CN102208023B (zh) * | 2011-01-23 | 2013-05-08 | 浙江大学 | 基于边缘信息和分布熵的视频字幕识别设计方法 |
| JP6095283B2 (ja) * | 2012-06-07 | 2017-03-15 | キヤノン株式会社 | 情報処理装置、およびその制御方法 |
| CN104769578B (zh) * | 2012-11-09 | 2017-06-27 | 加州理工学院 | 自动特征分析、比较和异常检测的方法 |
| CN104504401B (zh) * | 2015-01-09 | 2018-08-31 | 成都新舟锐视科技有限公司 | 一种基于多监控探头的目标识别系统 |
| CN104519326B (zh) * | 2015-01-09 | 2017-12-08 | 成都新舟锐视科技有限公司 | 一种联合多监控探头系统的目标跟踪方法 |
| JP6756338B2 (ja) | 2015-12-18 | 2020-09-16 | 株式会社リコー | 画像処理装置、画像処理システム、画像処理方法およびプログラム |
| US10469590B2 (en) * | 2018-01-02 | 2019-11-05 | Scanalytics, Inc. | System and method for smart building control using directional occupancy sensors |
| CN109389640A (zh) * | 2018-09-29 | 2019-02-26 | 北京字节跳动网络技术有限公司 | 图像处理方法和装置 |
| US10607105B1 (en) * | 2019-03-27 | 2020-03-31 | Disney Enterprises, Inc. | Perceptual data association |
| CN113379762A (zh) * | 2021-05-28 | 2021-09-10 | 上海商汤智能科技有限公司 | 图像分割方法、装置、电子设备和存储介质 |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH08287258A (ja) * | 1995-04-19 | 1996-11-01 | Fuji Xerox Co Ltd | カラー画像認識装置 |
| JPH08329247A (ja) * | 1995-05-30 | 1996-12-13 | Nippon Telegr & Teleph Corp <Ntt> | 動画像認識装置 |
| JP2000090277A (ja) * | 1998-09-10 | 2000-03-31 | Hitachi Denshi Ltd | 基準背景画像更新方法及び侵入物体検出方法並びに侵入物体検出装置 |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0363828B1 (en) | 1988-10-11 | 1999-01-07 | Kabushiki Kaisha Ouyo Keisoku Kenkyusho | Method and apparatus for adaptive learning type general purpose image measurement and recognition |
| JP2834153B2 (ja) | 1988-10-11 | 1998-12-09 | 工業技術院長 | 適応学習型汎用画像計測装置 |
| JP2982814B2 (ja) | 1988-10-11 | 1999-11-29 | 工業技術院長 | 適応学習型汎用画像計測方法 |
| JP4187448B2 (ja) * | 2002-03-07 | 2008-11-26 | 富士通マイクロエレクトロニクス株式会社 | 画像における移動体追跡方法及び装置 |
| CN100477745C (zh) * | 2002-08-09 | 2009-04-08 | 夏普株式会社 | 图像合成装置和图象合成方法 |
| JP4283005B2 (ja) | 2003-02-06 | 2009-06-24 | パナソニック株式会社 | 物体追跡方法、プログラム、及び記録媒体 |
| JP4090388B2 (ja) * | 2003-05-27 | 2008-05-28 | 三洋電機株式会社 | 電波受信装置 |
| CN100573548C (zh) * | 2004-04-15 | 2009-12-23 | 格斯图尔泰克股份有限公司 | 跟踪双手运动的方法和设备 |
| US7616782B2 (en) * | 2004-05-07 | 2009-11-10 | Intelliview Technologies Inc. | Mesh based frame processing and applications |
| US7376246B2 (en) * | 2005-06-27 | 2008-05-20 | Mitsubishi Electric Research Laboratories, Inc. | Subspace projection based non-rigid object tracking with particle filters |
-
2004
- 2004-12-02 JP JP2004349244A patent/JP3970877B2/ja not_active Expired - Fee Related
-
2005
- 2005-09-12 WO PCT/JP2005/016711 patent/WO2006059419A1/ja not_active Ceased
- 2005-09-12 EP EP05782365.0A patent/EP1835462A4/en not_active Withdrawn
- 2005-12-09 US US11/792,084 patent/US7957557B2/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH08287258A (ja) * | 1995-04-19 | 1996-11-01 | Fuji Xerox Co Ltd | カラー画像認識装置 |
| JPH08329247A (ja) * | 1995-05-30 | 1996-12-13 | Nippon Telegr & Teleph Corp <Ntt> | 動画像認識装置 |
| JP2000090277A (ja) * | 1998-09-10 | 2000-03-31 | Hitachi Denshi Ltd | 基準背景画像更新方法及び侵入物体検出方法並びに侵入物体検出装置 |
Non-Patent Citations (1)
| Title |
|---|
| See also references of EP1835462A4 * |
Also Published As
| Publication number | Publication date |
|---|---|
| EP1835462A4 (en) | 2014-11-12 |
| JP3970877B2 (ja) | 2007-09-05 |
| JP2006163452A (ja) | 2006-06-22 |
| EP1835462A1 (en) | 2007-09-19 |
| US7957557B2 (en) | 2011-06-07 |
| US20080187172A1 (en) | 2008-08-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP3970877B2 (ja) | 追跡装置および追跡方法 | |
| JP6482195B2 (ja) | 画像認識装置、画像認識方法及びプログラム | |
| CN112861808B (zh) | 动态手势识别方法、装置、计算机设备及可读存储介质 | |
| WO2006028106A1 (ja) | 異常動作検出装置および異常動作検出方法 | |
| Yadav et al. | Human illegal activity recognition based on deep learning techniques | |
| TW201832182A (zh) | 動作學習裝置、技能判別裝置以及技能判別系統 | |
| Alksasbeh et al. | Smart hand gestures recognition using K-NN based algorithm for video annotation purposes | |
| David | An intellectual individual performance abnormality discovery system in civic surroundings | |
| Devasena et al. | Video surveillance systems-a survey | |
| Kataoka et al. | Extended co-occurrence hog with dense trajectories for fine-grained activity recognition | |
| Lejmi et al. | A Novel Spatio-Temporal Violence Classification Framework Based on Material Derivative and LSTM Neural Network. | |
| Liang et al. | Three dimensional motion trail model for gesture recognition | |
| JP2007052609A (ja) | 手領域検出装置及び手領域検出方法、並びにプログラム | |
| Lin et al. | Smoking behavior detection based on hand trajectory tracking and mouth saturation changes | |
| Poonkodi et al. | RETRACTED ARTICLE: Action recognition using Correlation of Temporal Difference Frame (CTDF)—an algorithmic approach | |
| CN106682582A (zh) | 一种基于压缩感知外观模型的人脸跟踪方法及系统 | |
| WO2007112295A1 (en) | Method and apparatus for predicting the accuracy of a virtual scene based on incomplete information in video | |
| Haque et al. | Optical flow based violence detection from video footage using hybrid MobileNet and Bi-LSTM | |
| Kuppuswamy et al. | Similar Object Detection and Tracking in H. 264 Compressed Video Using Modified Local Self Similarity Descriptor and Particle Filtering. | |
| Du | Design of dance movement recognition algorithm based on 3d motion capture data | |
| Lei et al. | A novel approach for enhanced abnormal action recognition via coarse and precise detection stage | |
| Wensheng et al. | Implementation of virtual mouse based on machine vision | |
| Fernández et al. | Contextual and skin color region information for face and arms location | |
| Goudelis et al. | Identifying unintentional falls in action videos using the 3D Cylindrical Trace Transform | |
| Pandey et al. | VCFNet-A Multi-Stream Approach to Violence Detection in Complex Video Scenarios using Spatiotemporal Motion Fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| AK | Designated states |
Kind code of ref document: A1 Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BW BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE EG ES FI GB GD GE GH GM HR HU ID IL IN IS KE KG KM KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NA NG NI NO NZ OM PG PH PL PT RO RU SC SD SE SG SK SL SM SY TJ TM TN TR TT TZ UA UG US UZ VC VN YU ZA ZM ZW |
|
| AL | Designated countries for regional patents |
Kind code of ref document: A1 Designated state(s): BW GH GM KE LS MW MZ NA SD SL SZ TZ UG ZM ZW AM AZ BY KG KZ MD RU TJ TM AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IS IT LT LU LV MC NL PL PT RO SE SI SK TR BF BJ CF CG CI CM GA GN GQ GW ML MR NE SN TD TG |
|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2005782365 Country of ref document: EP |
|
| WWP | Wipo information: published in national office |
Ref document number: 2005782365 Country of ref document: EP |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 11792084 Country of ref document: US |