JPS5824732B2 - Pathway pattern recognition method for land mobile machines - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The safety of land mobile machines, such as automobiles, tractors, trains, etc., depends on their ability to detect obstacles in the path in which they travel.

In particular, when a robot is intended for movement, the ability to detect the distribution of obstacles in a passageway, that is, the ability to recognize passageway patterns, is an essential function.

Up until now, research and development from this perspective has been promoted in various countries around the world, but it has mainly been limited to the application of radar methods using microwaves, lasers, ultrasonic waves, etc.

However, this radar method has the drawback that its recognition ability in complex road environments is extremely inaccurate due to a reduction in resolution due to beam spreading.

Therefore, the present inventors
No. 8-2979), images of the external world are formed on two photoelectric conversion surfaces (such as television cameras) placed vertically apart on a mobile machine, and the image is electrically scanned to create a two-dimensional image. We proposed a method for converting the video pattern into a one-dimensional electrical signal and then recognizing the path pattern using a simple logic circuit.

With this method, much more precise image information can be obtained than with a single radar system, so if the processing method is devised, it may be possible to fairly accurately recognize the distribution of obstacles in a complex road environment. .

However, in the above method, in order to recognize a three-dimensional structure that exists at a certain distance by the presence of multiple brightness change points distributed in the height direction on the structure, those brightness change points are If the model pattern for obstacle determination is based on whether or not multiple corresponding pulses are generated, the illusion caused by the finite pulse width of the brightness change point or the complex pattern on the road surface, that is, the two-dimensional pattern is considered to be a three-dimensional obstacle. There was a risk that it would be recognized as

The illusion caused by the pulse width of the brightness change point can be solved by making the pulse width as narrow as possible, increasing the distance between the two television cameras, limiting the perceivable distance, etc.

On the other hand, the illusion caused by the planar pattern of the road surface is a problem involved in the recognition process itself, and cannot be solved so easily.

First, an example of the illusion caused by the planar pattern of the road surface in the above method will be specifically explained using the path pattern recognition device shown in FIG.

Now, when a mobile machine equipped with a path pattern recognition device as shown in Fig. 1 travels on a road surface marked with a black and white pattern with bright and dark change points at points A, B, and C, this device The video signal from the television camera is processed in the manner shown in FIG. 2 to recognize obstacles.

In other words, if the two television cameras in Figure 1 are both scanning vertically from top to bottom, the outputs a and b of each television camera are as shown in Figure 2. ,b
The video signal shown in is generated.

These video signals are converted into a signal C2d by a differentiating circuit at the next stage, and then converted into pulse sequences e and f having a constant amplitude and width by a waveform shaping circuit.

These correspond to the brightness/darkness change points A, B, and C of the black and white pattern in FIG.

Now, in order to match the light/dark change points corresponding to point A in the two pulse sequences e and f in Fig. 1, pulse sequence e is delayed like pulse sequence g using a shift register, and then the pulse sequence e is If it is applied together with the series f to the AND circuit at the next stage, a series of pulses will be generated as its output.

Then, by applying a pulse width corresponding to the height of the obstacle in the waveform shaping circuit at the next stage, the waveform shown in FIG. 2i can be obtained.

As a result, the final stage multiple pulse determination circuit generates the output shown in FIG. 2j, and this path pattern recognition device gives the illusion that the black and white pattern on the road surface is a three-dimensional obstacle.

The above example is the worst case, but on actual roads there are complex patterns such as trees and telephone poles, and the above illusion phenomenon occurs with a high probability.

Although the illusion of a two-dimensional pattern or the like as a three-dimensional obstacle is not a problem from a safety-first standpoint, it is not desirable for a recognition device.

In particular, when a path pattern recognition device is used to prevent collisions or automatically steer a vehicle, the above-mentioned illusion must be removed for the purpose of control.

The present invention aims to solve the problem of illusion in such a path pattern recognition method by improving its information processing method, and to remove the illusion phenomenon from the above method. The following recognition operation is added.

(1) Obstacle recognition is performed by classifying brightness change points into points changing from dark to bright and points changing from bright to dark in the scanning direction.

(2) When two or more bright/dark changing points exist in succession in the vertical scanning direction, it is recognized as an obstacle.

(3) When the presence of an obstacle is recognized in a plurality of adjacent vertical scans, it is confirmed as an obstacle for the first time.

Here, the operation (1) ensures recognition operation by roughly classifying the brightness change points into two types, reduces the illusion phenomenon explained in Figs. 1 and 2, and also (2)
) was established because the brightness change points of the same obstacle that occur in the upper and lower television cameras are similar, and, like the operation (1), it has the purpose of eliminating optical illusions. ing.

On the other hand, operation (3) attempts to determine the presence or absence of an obstacle from the recognition results obtained by multiple vertical scans, and the target obstacle must have a certain size in the left and right directions. is mentioned above.

By performing these operations, it is possible to almost eliminate optical illusions caused by road surface patterns and the like.

The recognition method of the present invention will be explained in more detail below with reference to FIG.

First, two photoelectric converters such as television cameras placed vertically apart on a land mobile machine convert a two-dimensional image of the target space into a one-dimensional electrical image signal, and convert the signal into a one-dimensional electrical image signal. Send it to the waveform shaping and delay circuit.

This "waveform shaping and delay" circuit is basically the same as the one shown in Figure 1, but the difference is that the brightness change point changes from dark to bright in the scanning direction, and from bright to dark. In order to roughly classify the points into two types: points that change to recognition” will be implemented.

The pulse sequence of the brightness change point group obtained from the video signal of one of the television cameras is delayed by a certain time by a digital delay circuit consisting of a shift register and a shift pulse generator, but this delay time depends on the bit capacity of the shift register. You can choose freely.

The two AND circuits A1 and A2 in the "vertical obstacle recognition" circuit perform mode-specific processing on the two types of brightness change point groups mentioned above. The AND circuit A2 calculates the AND of the brightness change point group (A) obtained from the video signal of the camera, and the logical product of the brightness change point group ←) in the AND circuit A2.

The pulse sequences of the light/dark change points of sweat) and H extracted by this change from dark to light and from bright to dark, which are distributed in the height direction at a distance corresponding to the above delay amount in the target space. This corresponds to the point where the brightness changes.

The above AND circuit AI, A2 is powered by three people because it uses two one-shot multis.
This is a measure to ensure the recognition operation by removing a certain period of time at the lower end from the recognition target.

The light/dark change points extracted by AND circuits A and A2 are then led to a 2-bit flip-flop via OR circuit 01 to AND circuit A1. Only when two temporally coincident brightness/darkness change points occur consecutively in A2, the output pulse of the OR circuit 01 based thereon causes the above-mentioned 2.
When the output of the bit flip-flop becomes "1" and there is a brightness change point that does not coincide in time between the output pulses from the OR circuit 01, a reset signal is applied from the AND circuit A3 via the OR circuit O6. The flip-flop is reset.

Next, in the "horizontal obstacle recognition" circuit, the output of the above flip-flop is led to AND circuit A, where two one-shot multis are used to recognize objects for a certain period of time at the left and right ends of the horizontal scan. The recognition operation is ensured by removing it from the

When the output of the flip-flop is "1" in two consecutive vertical scans, it is detected by the D-type flip-flop and the AND circuit A6 in the subsequent stage, and the AND circuit A
When the output of 6 becomes "1", this path pattern recognition device detects an obstacle.

Note that in the above recognition operation, recognition over the entire screen is completed in one period of horizontal scanning.

In this way, according to the path pattern recognition method of the present invention,
By using two television cameras and a simple logic circuit, it is possible to reliably recognize the distribution of obstacles in the path in the direction of movement of a moving machine, and the above-mentioned illusion can be greatly reduced, making it possible to It can be widely applied to collision prevention, vision for mobile robots, etc.

[Brief explanation of drawings]

Figure 1 is a block diagram of the previously proposed path pattern recognition device, and Figure 2 a=j is the output a = j of each part of the device.
FIG. 3 is a block diagram of a passage pattern recognition device embodying the present invention.

Claims (1)

[Claims] 1 Two-dimensional images of the target space are converted into one-dimensional electrical video signals by vertical scanning using two photoelectric converters arranged vertically apart, and the brightness and darkness of the target space obtained from these video signals is calculated. The pulse sequence corresponding to the changing point is separated into two types of pulse sequences with bright and dark changing points that change from dark to bright and from bright to dark, and the pulse sequence of the bright and dark changing points obtained from one video signal is separated for a certain period of time. The pulse sequences of the two types of brightness change points obtained from both video signals are logically ANDed using an AND circuit, and two or more of the brightness change points extracted by these are successively detected. Path pattern recognition method for land mobile machines, which is characterized by sometimes recognizing obstacles in the target space. Each is converted into one-dimensional electrical video signals by vertical scanning in a photoelectric converter, and pulse sequences corresponding to brightness change points in the target space obtained from these video signals are converted into two pulse sequences that change from dark to bright and from bright to dark. In addition to separating the pulse series of the light and dark change points of the species, the pulse series of the light and dark change points obtained from one video signal is delayed by a certain period of time, and the pulses of the two types of light and dark change points obtained from both video signals are separated. A logical product is performed for each series using an AND circuit, and when two or more of the brightness change points extracted by these occur consecutively, and when they occur consecutively in a plurality of adjacent vertical scans, A path pattern recognition method for a land mobile machine characterized by recognizing an obstacle in a target space.