JP4509429B2 - Obstacle recognition method and recognition processing apparatus using a stereo camera - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an obstacle recognition method using a stereo camera and a recognition processing apparatus thereof, and more particularly to an obstacle recognition method using a stereo camera that searches for a pedestrian image or the like using a difference in the angle of view of left and right cameras and the recognition thereof. The present invention relates to a processing apparatus.
[0002]
[Prior art]
In general, an obstacle recognition apparatus using an in-vehicle camera is required to recognize not only a white line on a road but also an obstacle such as a preceding vehicle or a pedestrian.
[0003]
However, unlike pedestrian images, images such as pedestrians are divided into multiple parts even after preprocessing such as density conversion processing, noise removal filter processing, differentiation processing, grouping processing, and labeling processing. There are many. Moreover, it is difficult to correlate the images that have been divided into a plurality of images, and it is difficult to recognize obstacles themselves.
[0004]
FIG. 6 is a schematic diagram showing the positional relationship between a pedestrian and an automobile when a monocular in-vehicle camera for explaining an example of the prior art is used. As shown in FIG. 6, when a conventional monocular (single) in-vehicle camera captures a pedestrian 20 (for example, a position of 10 meters: m) that actually exists in a real space position, the pedestrian image is displayed. Since it is divided into a plurality of parts such as a head, a torso, and a foot, the pedestrian 20 recognizes that the pedestrian 20 was present at an incorrect position 21 in the real space (for example, a position of 40 m) from the angle relationship. It is easy to end up.
[0005]
FIG. 7 is a perspective image obtained by the monocular on-vehicle camera used in FIG. As shown in FIG. 7, when this monocular on-vehicle camera captures the above-described pedestrian 20 or the like, the torso portion is in spite of the fact that the foot portion 20B is at an actual position, for example, 10 m, due to perspective. The illusion is that 20A is at the position of the white line (thick solid line), that is, at the position of 40 m. In short, in this case, an image extracted through preprocessing is often divided into a head, a torso 20A, a leg, a foot 20B, and the like.
[0006]
Further, FIG. 8 is a schematic diagram showing the positional relationship between a pedestrian and a car when using a stereo in-vehicle camera for explaining another conventional example. As shown in FIG. 8, the conventional stereo vehicle-mounted camera exists at a position corresponding to the image of the foot 23 in the real space from the left camera line of sight and a position corresponding to the image of the torso 24 in the real space from the left camera line of sight. When a pedestrian (for example, a position of 10 m) is photographed, the image of the torso 24 is assumed to be grounded in real space from the left camera line of sight (for example, 40 m) It is easy to recognize it as a position.
[0007]
FIG. 9 is a perspective image obtained by the left camera of the stereo in-vehicle camera used in FIG. As shown in FIG. 9, when an image is taken with the left camera, an image of a pedestrian's foot (such as shoes) 26 exists, for example, at a point of 10 m from the camera angle and perspective relationship, and other than the foot (such as a torso). For example, the image 27 is recognized as existing at a point of a white line of 40 m.
[0008]
FIG. 10 is a perspective image obtained by the right camera of the stereo in-vehicle camera used in FIG. As shown in FIG. 10, when photographing with the right camera, the image of the foot 28 corresponding to the right camera is present at a point of, for example, 10 m from the camera angle and perspective relationship, and other than the foot of the pedestrian (such as the torso) ) The 29 image is recognized as being present at a point of 30 m, for example. In addition, the image of the trunk portion 30 does not exist at the location corresponding to the erroneous position 25 in FIG.
[0009]
Thus, the images obtained by the stereo in-vehicle camera are also divided into the head, the torso (27, 29), the legs, the feet (26, 28), etc., as shown in FIGS. There are many.
[0010]
The obstacle recognition apparatus described above must connect these disjoint image groups based on a certain rule and recognize a pedestrian or the like as an obstacle. In such a conventional technique, as a recognition method, images in the vicinity are regarded as one set and specified by pattern matching as to whether or not the image is an obstacle image of a target pedestrian or the like.
[0011]
[Problems to be solved by the invention]
However, the conventional recognition method or recognition device described above may not be able to make a reliable determination because it is difficult to identify a pedestrian or the like as an obstacle even by pattern matching when there are excesses or deficiencies in the connected images. .
[0012]
Even if it can be partially determined, in the single camera (FIG. 7), when the image of the foot 20B or the like is lost at the preprocessing stage, the body 20A and the foot 20B are connected. Since this is not possible, the processing is performed assuming that the trunk portion 20A is grounded alone and on the road surface. As a result, there is a possibility that a thing in the real space position (the position of the pedestrian 20 in FIG. 6) in the road surface coordinate system of the trunk portion 20 </ b> A is misunderstood as being in the wrong position 21.
[0013]
Further, even in the stereo camera, when the image of the foot 26 or the like is lost as a result of the preprocessing in FIG. 9, the left camera image cannot be connected to the body 27 and the foot 26. It is necessary to search the body 29 of the right camera image with strict matching. If that fails, the processing is performed assuming that the trunk portion 29 is alone and is in contact with the road surface. As a result, in the road surface coordinate system of the torso 29, what should be at the position in the real space (the pedestrian 24 in FIG. 8) despite the absence of the torso image 30 by the right camera for the wrong position 25 in FIG. There is a possibility of misunderstanding that it is in the wrong position 25.
[0014]
Therefore, in the single camera system, since it is impossible to calculate the road surface coordinates for a single image floating on the road surface, it is necessary to reliably extract an image such as a foot part touching the road surface.
[0015]
Further, in the case of a stereo camera in which images with different distances exist adjacent to each other, since it is necessary to make a determination after combining images from both cameras, there is a drawback that the processing becomes complicated and a reliable determination cannot be made.
[0016]
In order to improve the determination accuracy, a matching pattern library for determining the shape of an infinite obstacle is required, and a processing apparatus for processing the matching pattern library at high speed is also required.
[0017]
An object of the present invention is to provide a method of recognizing an obstacle by a stereo camera and a recognition processing device thereof that make use of the features of a stereo camera, eliminate the need for a general-purpose matching pattern library, and do not require a high processing speed. There is to do.
[0018]
[Means for Solving the Problems]
In the obstacle recognition method using the stereo camera according to the present invention, the image preprocessing unit connected to the stereo camera performs filtering, binarization, and differentiation on the left and right images obtained from the stereo camera. A camera image preprocessing step, a labeling processing step for performing a labeling process on the differential image of the left and right cameras using a labeling processing unit with respect to an output of the image preprocessing unit, and a left and a right non-componentized by the labeling processing unit A search processing step for performing a search process on a connected image using a search processing unit, and a real space image and its size for a group image obtained using an obstacle coordinate output coordinated by a coordinate conversion unit. Using the analysis step for analyzing the image and the size of the image obtained from the analysis result of the analysis step in the image processing / control device. An obstacle determination step for determining whether the vehicle is a two-wheeled vehicle or other obstacle, and controls the acceleration / deceleration of the vehicle by controlling the brake / acceleration control device according to the output of the obstacle determination step, and at the same time, The steering device is configured to control the steering amount of the steering wheel.
[0019]
Further, the search process in the present invention includes a search area setting step of searching for an obstacle using the forward road surface area in the straight traveling state calculated from the camera angle of view as a temporary search area, and the search area from the front in the left camera image. The left camera candidate image search step for searching and obtaining the first candidate connected image, and when there is the candidate image, there is a paired image at the corresponding position of the right camera image assuming that the candidate image is grounded A right camera pair image search step for searching whether or not to determine whether the left and right candidate connected image pairs obtained by the left camera candidate image search step and the right camera pair image search step match When there is a matching image in the matching image determination step and the matching image determination step, the paired images are recorded, and the group image A group image search processing step for performing a search process, and then a next candidate search step for searching for a candidate connected image as the next ground connected image in the search region of the left camera image when the group image search process ends. And when it is determined that there is no candidate image in the iterative process, the search process is terminated.
[0020]
Further, the group image search processing according to the present invention includes a left camera contact image search step for searching for a contact image existing around a coincidence connected image obtained from an image of the left camera and within a predetermined distance. And when there is a contact image in the left camera contact image search step, a right camera contact pair image search step for searching whether there is a contact image having the same positional relationship as the left camera image in the matching connection image of the right camera image; As a result of searching for the paired images, a registration step of registering the target image as an image group, and then a next image searching step of searching for the next image in contact with the candidate image in the left camera image. And when it is determined that the contact image does not exist in the repetitive processing, the group image search processing is ended.
[0021]
Also, the obstacle recognition method using the stereo camera according to the present invention obtains the road surface coordinates on the three-dimensional space coordinates calculated from the camera angle of view of the stereo camera, and searches for a connected image grounded to the road surface from the road surface coordinates. A search step for performing matching, a pairing processing step for determining matching between the left and right connected images, and a whole image searching step for searching for a whole image of obstacle images dispersed around the matching connected images in the pairing processing step. And the number of candidates for other search processing target linked images is reduced by the overall image search step.
[0022]
Furthermore, the obstacle recognition processing device of the present invention includes a stereo camera that captures the traveling direction of the vehicle at an arbitrary angle, and an image processing unit that processes an image output of the stereo camera, and the image processing unit includes: The stereo camera image output is input, and the output of the pair of image preprocessing units including the filter processing function, the binarization processing function, and the differentiation circuit for the image is input. A pair of labeling processing units for performing a labeling process for creating a connected image based on the differential image, and a search processing unit for inputting a output of the pair of labeling processing unit and searching for a specific image with respect to the connected image And a coordinate conversion processing unit that converts the image searched by the search processing unit into a road surface coordinate system.
[0023]
Further, the image processing unit in the obstacle recognition processing device of the present invention supplies the obstacle coordinate output output from the coordinate conversion unit to the image processing / control device in the vehicle, and the brake / acceleration control device, It is configured to control the power steering device.
[0024]
DETAILED DESCRIPTION OF THE INVENTION
The embodiment of the present invention makes use of the features of a stereo camera, uses road surface coordinates on a three-dimensional space determined from left and right camera angles, and uses ground connection images in a narrow range as a reference for search. By recursively using a simple process of connecting images that are in contact with it, an obstacle recognition process that eliminates the need for a general-purpose matching library and does not require a very high processing speed. It is to be realized.
[0025]
Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram of a stereo camera and an image processing unit for explaining an embodiment of the present invention. As shown in FIG. 1, the recognition processing apparatus for recognizing an obstacle in the present embodiment is a twin-lens (stereo) camera including a right camera R and a left camera L that captures the traveling direction of the vehicle at an arbitrary angle. 1 and an image processing unit 7 for processing the image output of the stereo camera 1. The image processing unit 7 inputs the image output of the right camera R and the left camera L in the stereo camera 1 respectively, and a pair of image preprocessing units 2 having a filter processing function, a binarization processing function, and a differentiation circuit for the image. And a pair of labeling processing units 3 for inputting the outputs of the pair of image preprocessing units 2 and applying a labeling process for creating a connected image based on the differential image, and outputs of the pair of labeling processing units 3 And a search processing unit 4 for searching for a specific image with respect to the obtained connected image, and a coordinate conversion processing unit 5 for converting the image searched by the search processing unit 4 into a road surface coordinate system, The obstacle coordinate output 6 of the coordinate conversion processing unit 5 is supplied to an image processing / control device to be described later.
[0026]
FIG. 2 is a cross-sectional view of the front half of an automobile equipped with an image processing / control device including the image processing unit shown in FIG. As shown in FIG. 2, in such an automobile, the image processing unit 7 described with reference to FIG. 1 is mounted, the output of the stereo camera 1 is processed, and an image processing / control device 8 for instructing each internal device is provided. And a brake / accelerator control device 9 for controlling acceleration / deceleration of the vehicle, and an electric power steering device 10 for controlling the steering amount, respectively, receiving control signals from the image processing / control device 8. In addition, the image processing unit 7 is illustrated as being integrated with the stereo camera 1 here, but may be separated from the stereo camera 1 and integrated with the image processing / control device 8 side.
[0027]
Next, the operation of the above-described recognition processing apparatus will be described with reference to the positional relationship between FIGS. 1 and 2 and FIGS.
[0028]
FIG. 3 is a processing flowchart of the image processing / control apparatus in FIG. As shown in FIG. 3, first, in the left and right camera image preprocessing step S1, the image preprocessing unit 2 performs filtering, binarization, and differentiation on the left and right images obtained in advance from the stereo camera 1. Perform pre-processing. Next, in the labeling processing step S2, the labeling processing unit 3 is used to perform the labeling processing on the differential images of the left and right cameras. Thereafter, the search processing unit 4 performs search processing on the left and right connected images that have been made non-partici- ated by these processes in search processing step S3.
[0029]
Next, in analysis step S4, a real space image, its size, etc. are analyzed with respect to the group image calculated | required using the obstacle coordinate output 6 coordinated by the coordinate transformation part 5. FIG. Further, in the obstacle determination step S5, the image processing / control device 8 determines whether the object is a pedestrian, a two-wheeled vehicle, or other obstacles according to the size of the image obtained from the analysis result.
[0030]
Based on such processing and determination, the image processing / control device 8 controls the brake / accelerator control device 9 to control the acceleration / deceleration of the vehicle, and simultaneously controls the electric power steering device 10 to control the steering amount of the steering wheel. Control.
[0031]
The detailed operation of the search process S3 described above is as follows.
[0032]
FIG. 4 is a flowchart of the search process in FIG. As shown in FIG. 4, in the search process (step S3), first, in the search area setting step S6, the area on which the obstacle must be searched is calculated on the three-dimensional space coordinates calculated from the fixed camera angle of view. The forward road surface area in the straight traveling state is set as a temporary search area. In this case, instead of the front road surface area in the straight traveling state, the front road surface area can be obtained by another white line recognition process or the like, and this can be used as the search area. In any case, when this temporary search area is set, in the left camera candidate image search step S7, the search area is searched from the front (the lower side of the screen) with the left camera image, and the foot as the first candidate connected image is displayed. The unit 26 (see FIG. 9) is obtained.
[0033]
Next, when it is searched that there is a candidate image in step S7 described above, a repetition process (while: there is a candidate image) is performed in repetition step S8 (if there is no candidate image, return). That is, when there is a candidate image in this repeat step S8, first, in the right camera corresponding image search step S9, it is assumed that the candidate image is in contact with the road surface, the road surface coordinates are calculated, and the right camera image corresponds. The foot portion 28 (see FIG. 10) as a connected image of the road surface coordinate position is searched.
[0034]
Further, in the coincidence image determination step S10, it is determined whether or not the foot portions 26 and 28 as the left and right candidate connection image pairs obtained in the above-described steps S7 and S9 are coincident connection images. For example, as specific determination means, the values of the connected images in the X and Y directions, the total number of pixels, and the like are compared, and the right connected image is within ± 10% of the value of the left connected image. If the value of is entered, there is a method to consider that they match. In short, the method of confirming that one is used as a reference and the other is within a certain range is the simplest and most efficient. As other determination means, a correlation value is obtained for the left and right connected images using a correlation function, and whether or not the correlation value is equal to or larger than a certain value is compared, or a feature of the connected image is subjected to skeleton processing. A method for obtaining and comparing the correlation values can be considered.
[0035]
As described above, for the feet 26 and 28 as the matched connected image pair determined to be matched in the matched image determining step S10, the pair grounded image is stored in the search result database (not shown) in the recording step S11. At the same time, the road surface coordinates are registered. Further, when there is a coincidence image, in group image search processing step S12, a group image search process that will be separated into parts in the vicinity is performed. On the other hand, in the coincidence image determination step S10, when there is no or no coincidence image, the process proceeds to the next step S13.
[0036]
Finally, when the group image search processing in the group image search processing step S12 is completed, in the next candidate search step S13, a candidate connected image as a next ground connected image (however, the search described above is performed) in the search area of the left camera image. The connected images registered in the result database are excluded from candidates). If there is no next candidate in step S13, the process returns to return and the search processing step S3 is terminated.
[0037]
FIG. 5 is a flowchart of the group image search process in FIG. As shown in FIG. 5, in the group image search processing step (S12: see FIG. 4), in the left camera contact image search step S14, around the foot part 26 as a matched connected image obtained from the image of the left camera and It is searched whether there is a connected image (hereinafter referred to as a contact image) existing within a certain distance (number of blank pixels). Here, the repeating step S15 repeats the following processing while there is a contact image.
[0038]
Next, when there is a contact image 26 in the repeat step S15, whether there is a contact image having the same positional relationship as the left camera image in the matching connection image of the right camera image in the pair image search step S16 with respect to the contact image 26 or the like. Search for no. Next, as a result of the search for the paired images, the matching determination process similar to step S14 in FIG. 4 described above is performed in the target image determination step S17 on the left and right contact image pairs 26, 29 and the like.
[0039]
Thereafter, if it is determined in the target image determination step S17 that the target images are present and coincide with each other, they are stored in the search result database as child connected images of the ground connected image pairs 26 and 28 previously registered in the registration step S18. Register an image group. The target image determination step S17 is continued until there are no more target images.
[0040]
Further, when the target image disappears in step S17, in next candidate search step S19, a search is made as to whether or not there is another contact image such as a head on the basis of the group image. That is, the next image in contact with the candidate image is searched for in the left camera image. In short, iterative processing from step S15 to step S19 is performed on all contact images.
[0041]
When these processes are performed on all such candidate connected images and all the ground connected image pairs and group image pairs are searched, the processing of the candidate images in step S8 of FIG. 4 described above is completed.
[0042]
According to the present embodiment described above, the road surface coordinates in the three-dimensional space calculated from the left and right camera angles are used, and the ground connection image such as the foot is used as a reference for the search. Is mentioned.
(A) The search range can be set narrow in advance, and unnecessary search processing is not required.
(B) It is only necessary to determine the coincidence of the left and right connected images by a simple method, and it does not require a huge library or ultra-high speed processing required for pattern matching processing, etc. No judgment error occurs.
[0043]
Further, in the present embodiment, the following advantages can be cited when searching for a contact image (group image) with respect to the coincidence connected image.
(A) Since the search range can be limited, useless search processing is not required.
(B) Similarly, since it is only necessary to determine the coincidence of the left and right connected images by a simple method, there is no need for an enormous library required for pattern matching processing or the like, or in the left and right camera images. No misjudgment for minor feature differences.
(C) By the group image determination process, it is possible to remove the target from the search candidates for the next ground connection image and contact connection image, and the overall processing can be speeded up.
(D) Regardless of the overall image of the obstacle, it is possible to search for the presence or absence of the obstacle, and there is room for expansion of the determination of whether the obstacle is a pedestrian, bicycle, or other stationary object It is possible to give
[0044]
Although one embodiment of the present invention has been described above, other modifications may be made as follows.
[0045]
The reference image in the stereo camera described above may be based on either the left or right camera image. Further, as determination means for determining whether or not the pair of left and right candidate connected images is a matched connected image, the values of the connected images in the X direction, the Y direction, the total number of pixels, and the like are compared, and the left If the value of the right connected image is within ± 10% of the value of the connected image, it may be considered that they match.
[0046]
Furthermore, a correlation value is obtained for the left and right connected images using a correlation function, a method for comparing whether or not the correlation value is equal to or greater than a certain value, and a method for obtaining and comparing features of the connected images by performing skeleton processing or the like May be adopted.
[0047]
In addition, as a method for searching for a contact image in the above-described embodiment, a determination criterion of a certain distance (the number of blank pixels) or less is provided, or the center of gravity of a connected image or a contact side [a certain distance (the number of blank pixels) ) The length of the following gap] is also possible as a criterion.
[0048]
In the above-described embodiment, the electric power steering device has been described as an example in order to control the steering amount of the steering wheel. However, the power steering device is controlled using hydraulic means in addition to being electrically controlled. You can also
[0049]
【The invention's effect】
As described above, the obstacle recognition method using the stereo camera and the recognition processing apparatus according to the present invention use the road surface coordinates in the three-dimensional space determined from the left and right camera angles, and the connected image is grounded to the road surface. Under the assumption that the other camera image is used, the other camera image is searched for the same road surface coordinates, the image actually touching the road surface is extracted, and the connected image in contact with the connected image is extracted. It is possible to extract the whole image of obstacles such as pedestrians and obtain the road surface coordinates at the same time, eliminating the need for a general-purpose matching pattern library and accurate processing without requiring a high processing speed. There is an effect that obstacle determination can be realized. In addition, according to the present invention, it is possible to extend the determination of whether the obstacle is a pedestrian, a bicycle, or other stationary objects by obtaining the characteristics of the extracted whole image of the obstacle.
[Brief description of the drawings]
FIG. 1 is a block configuration diagram of a stereo camera and an image processing unit for explaining an embodiment of the present invention.
FIG. 2 is a cross-sectional view of the front half of an automobile equipped with an image processing / control device including the image processing section shown in FIG.
3 is a process flow diagram of the image processing / control apparatus in FIG. 2. FIG.
4 is a flowchart of the search process in FIG. 3. FIG.
FIG. 5 is a flowchart of group image search processing in FIG. 4;
FIG. 6 is a schematic diagram showing the positional relationship between a pedestrian and a car when using a monocular on-vehicle camera for explaining an example of the prior art.
7 is a perspective image obtained with the monocular on-vehicle camera used in FIG. 6. FIG.
FIG. 8 is a schematic diagram showing the positional relationship between a pedestrian and a car when a stereo in-vehicle camera for explaining another conventional example is used.
9 is a perspective image obtained by the left camera of the stereo in-vehicle camera used in FIG.
10 is a perspective image obtained by the right camera of the stereo vehicle-mounted camera used in FIG.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Stereo camera 2 Image pre-processing part 3 Labeling process part 4 Search process part 5 Coordinate conversion part 6 Obstacle coordinate output 7 Image process part 8 Image processing and control apparatus 9 Brake and accelerator control apparatus 10 Electric power steering apparatus S3 Search process step S12 Group image search processing step

Claims (6)

A camera image that generates a differential image by performing filter processing, binarization processing, and differentiation processing on the left and right images obtained from the left and right cameras included in the stereo camera by an image preprocessing unit connected to the stereo camera A pre-processing step;
Using a labeling processing unit, a labeling process step for performing a labeling process for generating a connected image based on the differential image;
A search processing step of inputting the output subjected to the labeling process, performing a search process on the connected images using a search processing unit, and obtaining a group image from the specific connected images;
Using the obstacle coordinate output converted to the road surface coordinate system by the coordinate conversion processing unit, the analysis step of analyzing the real space image and its size for the obtained group image,
An obstacle determination step for determining whether the image processing / control device uses the size of the image obtained from the analysis result of the analysis step and determines whether the object is a pedestrian, a two-wheeled vehicle, or other obstacles,
According to the stereo camera, the brake / accelerator control device is controlled by the output of the obstacle determination step to control acceleration / deceleration of the vehicle, and at the same time, the steering amount of the steering wheel is controlled by controlling the power steering device. Obstacle recognition method. The search process includes
A search area setting step for searching for an obstacle using a forward road surface area in a straight traveling state calculated from the angle of view of the camera as a temporary search area;
A left camera candidate image search step for searching the search area from the front with a left camera image output by the left camera and obtaining a first candidate connected image;
When there is the candidate image, assuming that the candidate image is grounded, a right camera pair image that searches for a paired image at the corresponding position of the right camera image output by the right camera A search step;
A matching image determining step for determining whether the left and right candidate connected image pairs obtained by the left camera candidate image searching step and the right camera pair image searching step are the same connected images;
When there is a matching image in the matching image determination step, a group image search processing step of recording the paired images and performing a search process of the group image;
Thereafter, when the search process of the group image is completed, a next candidate search step of searching a candidate connected image as a next ground connected image in the search area of the left camera image,
The obstacle recognition method using a stereo camera according to claim 1, wherein when the iterative process determines that there is no candidate image, the search process is terminated. The group image search process includes:
A left camera contact image search step for searching for a contact image that exists around the coincidence connected image obtained from the left camera image and within a predetermined distance; and
When there is a contact image in the left camera contact image search step, a right camera contact pair image search step for searching whether there is a contact image having the same positional relationship as the left camera image in the matched connection image of the right camera image; ,
As a result of this pair image search, a registration step of registering the target image as the image group;
Thereafter, a next image search step of searching for a next image in contact with the candidate image in the left camera image,
The obstacle recognition method using a stereo camera according to claim 2, wherein when it is determined that there is no contact image in the iterative process, the group image search process is terminated. A search step for obtaining road surface coordinates on a three-dimensional space coordinate determined from a camera angle of view of a stereo camera, and searching for a connected image grounded to the road surface from the road surface coordinates;
A pairing process step for determining matching of the left and right connected images;
An overall image search step for searching for an overall image of obstacle images dispersed around the matching connected image in the pairing processing step,
An obstacle recognition method using a stereo camera, wherein the overall image search step reduces the number of candidates for other search processing target linked images. A stereo camera that captures the traveling direction of the vehicle at an arbitrary angle;
An image processing unit for processing image output of the stereo camera;
The image processing unit
A pair of image pre-processing units that respectively input image outputs of the stereo cameras, and include a filter processing function, a binarization processing function, and a differentiation circuit for the images;
A pair of labeling processing units that respectively input outputs of the pair of image preprocessing units and perform a labeling process for creating a connected image based on the differential image;
A search processing unit that inputs outputs of the pair of labeling processing units and searches for a specific image with respect to the connected images;
An obstacle recognition processing apparatus comprising: a coordinate conversion processing unit that converts an image searched by the search processing unit into a road surface coordinate system. The image processing unit
6. The obstacle recognition processing device according to claim 5, wherein an obstacle coordinate output outputted from the coordinate conversion processing unit is supplied to an image processing / control device in the vehicle to control a brake / acceleration control device and a power steering device. .

Images