JP4155252B2 - Vehicle detection device - Google Patents

Description

  The present invention relates to an apparatus for detecting a vehicle existing around a host vehicle.

  There is known an apparatus for detecting a vehicle existing around a host vehicle using a captured image of a camera. In such an apparatus, it is necessary to enable stable detection even when the brightness or shape of a vehicle to be detected changes. In the apparatus disclosed in Patent Document 1, the detection target pattern is updated using the detection result, and the next pattern matching is performed to perform stable vehicle detection against changes in brightness and shape. .

JP-A-6-30417

In the device disclosed in Patent Literature 1, a vehicle detection method other than pattern matching cannot be selected. Thus, there is a need for a vehicle detection device that can perform stable vehicle detection by selecting an appropriate vehicle detection method according to the situation.

  The vehicle detection device according to the present invention includes an imaging unit that acquires a captured image around the host vehicle, and candidate points as candidate feature points representing a part of another vehicle existing around the host vehicle in the captured image from the captured image. A candidate point extracting means for extracting; a selecting means for selecting a vehicle detection method for detecting other vehicles based on the position or moving speed of the candidate point extracted by the candidate point extracting means; and Vehicle detection means for detecting another vehicle using the vehicle detection method selected by the selection means.

  According to the present invention, a vehicle detection method is selected, and other vehicles around the host vehicle are detected using the selected vehicle detection method based on the captured image. Since it did in this way, the suitable vehicle detection method can be selected according to a condition, and the stable vehicle detection can be performed.

  A configuration of a vehicle detection apparatus according to an embodiment of the present invention is shown in FIG. This vehicle detection device is mounted on a vehicle and used to notify the driver of the presence of the other vehicle behind the host vehicle by detecting the other vehicle from a captured image obtained by imaging the rear of the host vehicle. It is. The vehicle detection device includes a camera 1, an image memory 2, a microcomputer 3, and a monitor 4.

  The camera 1 is installed at the rear of the host vehicle so as to capture the rear of the host vehicle, and outputs the captured image to the image memory 2 as an image signal. As the camera 1, for example, a CCD camera or the like is used. The image memory 2 is a frame memory for storing and holding an image signal input from the camera 1 as image data. The image data stored in the image memory 2 is output to the microcomputer 3 in units of image frames under the control of the microcomputer 3.

  The microcomputer 3 performs processing as will be described later based on the image data from the image memory 2, thereby extracting another vehicle from the captured image, and the distance and speed of the other vehicle in the real space with respect to the own vehicle. Is calculated. When the calculated distance or speed satisfies a predetermined condition, the monitor 4 is controlled by the microcomputer 3 and a marker for indicating the presence of an obstacle is displayed on the monitor 4. The monitor 4 is installed in an easy-to-view place near the driver's seat, and displays a marker to make the driver recognize the other vehicle as an obstacle. In this way, the driver of the host vehicle is notified of the presence of another vehicle traveling behind the host vehicle.

  A method for extracting another vehicle from the captured image in the microcomputer 3 will be described. At this time, the microcomputer 3 executes new vehicle detection and vehicle tracking processes. The new vehicle detection process is a process for extracting other vehicles newly captured in the captured image, and the vehicle tracking process is for tracking and extracting other vehicles already captured in the captured image. It is processing. That is, each other vehicle is first extracted from the captured image by the new vehicle detection process, and thereafter extracted from the captured image by the vehicle tracking process.

  The contents of the new vehicle detection process will be described before. In the new vehicle detection process, a point representing the characteristics of the vehicle and representing a part of another vehicle is detected from the edge intersections extracted from the captured image. Such an edge intersection is hereinafter referred to as a feature point. At this time, according to the position of the target edge intersection point on the screen, one of two types of feature point detection methods described later is selected and used.

  In addition, the extraction of the edge intersection from the captured image is performed by extracting the vertical edge (vertical edge) and the horizontal edge (horizontal edge) from the captured image and obtaining the intersection of the vertical edge and the horizontal edge. . Of the edge intersections thus extracted as feature point candidates, an edge intersection that actually represents a part of another vehicle is detected as a feature point. The edge here is a portion indicating the boundary line of the region occupied by the object in the captured image. For this edge extraction, an image processing method using a known image filter called a Sobel filter can be used.

  Pattern matching processing is used as a feature point detection method for edge intersections located above a predetermined position on the screen, that is, edge intersections whose distance from the vehicle in the real space is longer than the predetermined distance. In this pattern matching process, an image pattern within a predetermined range around each edge intersection is compared with an image pattern stored in advance as a basic vehicle pattern, and the comparison result satisfies a predetermined pattern matching condition. The corresponding edge intersection is detected for each vehicle as a feature point.

  On the other hand, a grouping process is used as a feature point detection method for an edge intersection located below a predetermined position on the screen, that is, an edge intersection whose distance from the host vehicle is closer than a predetermined distance in real space. . In this grouping process, the edge intersections moving in almost the same direction and speed on the screen, that is, the edge intersections in which the difference in the velocity vector in the captured image is equal to or smaller than a predetermined value are grouped, and the corresponding edge intersection is characterized. A point is detected for each vehicle. At this time, a predetermined value to be set for the speed vector difference can be determined in advance. In order to calculate the velocity vector of the edge intersection, the edge intersection needs to be extracted over at least two frames.

  When a feature point is detected by the pattern matching process or the grouping process, an area corresponding to each other vehicle (hereinafter referred to as a vehicle area) is set in the captured image based on the position of each feature point on the screen. Furthermore, the arrangement of feature points and the vehicle area set based on the feature points are registered as a vehicle pattern for each vehicle. Unlike the basic vehicle pattern described above, this vehicle pattern is determined only by the arrangement of feature points and the size of the vehicle area, not the pattern of the image. As described above, the new vehicle detection process is performed.

  Next, the contents of the vehicle tracking process will be described. The vehicle tracking process is performed on edge intersections that have already been detected as feature points in the previous image frame among edge intersections extracted from the captured image. At this time, in the vehicle tracking process, one of the following two detection methods is selected and used according to the moving speed of the edge intersection in the captured image. When the movement speed of the edge intersection is slow, a detection method as shown in FIG. 2 is used, and when the movement speed is fast, a detection method as shown in FIG. 3 is used. Hereinafter, the detection method shown in FIG. 2 will be described.

  FIG. 2A shows a captured image acquired by the camera 1 at time T1. Further, (b) shows a captured image acquired by the camera 1 at the time of the next image frame after time T1 (referred to as time T2). In these captured images, the other vehicle 100 approaching from behind the host vehicle is captured. It is assumed that edge intersections 21a, 21b, 21c and 21d and edge intersections 22a, 22b, 22c and 22d of the other vehicle 100 are extracted from the captured images of (a) and (b), respectively. In addition, as for each of these edge intersections, those representing the same point in the vehicle are associated with each other. For example, the edge intersection 21a in (a) is associated with the edge intersection 22a in (b).

  By calculating the position change amount between the edge intersections associated with each other as described above, the moving speed in the captured image of each edge intersection at time T2 is obtained. For example, the moving speed of the edge intersection 22a at time T2 is obtained from the amount of change in position from the edge intersection 21a to the edge intersection 22a. And the average value of the moving speed is calculated about the edge intersection showing the same vehicle, ie, edge intersection 22a-22d. Here, it is assumed that the average value of the moving speeds of the edge intersections 22a to 22d is equal to or less than a predetermined value.

  Here, it is assumed that the edge intersections 21a to 21d of the captured image of (a) are detected as feature points of the other vehicle 100 at time T1. In addition, it is assumed that a rectangular area that is located at the center of these feature points 21 a to 21 d and that can include each feature point is set as the vehicle area 11. At time T1, the feature points 21a to 21d and the vehicle area 11 are registered as the vehicle pattern of the other vehicle 100 as shown in (c).

  As described above, when the average value of the moving speeds of the edge intersections at time T2 is equal to or less than the predetermined value, the vehicle including the probability distributions 32a, 32b, 32c and 32d shown in (d) based on the vehicle pattern shown in (c). Calculate pattern presence distribution. The probability distributions 32a to 32d are obtained as follows. First, the predicted position of the feature point at time T2 is obtained based on the relative position and moving speed of the feature points 21a to 21d at time T1. Note that the predicted position obtained here determines the relative positional relationship between the feature points. Next, the relative existence probability of each feature point at time T2 is set for the predicted position. At this time, the highest existence probability is set in the central portion of the predicted position, and a lower existence probability is set around it. In this way, the probability distributions 32a to 32d are obtained by setting the existence probabilities for the predicted positions. The probability distributions 32a to 32d are set so as to define the relationship between the relative positions and the existence probabilities, and the positions are not fixed with respect to the captured image.

  The probability distributions 32a to 32d in (d) illustrate that the existence probability of the feature point is high in a certain range of the filled central portion, and the existence probability of the feature point is low in the portion indicated by hatching around it. . The setting value of the existence probability and the size of the setting range in each of the central part and the peripheral part are determined in advance. Note that, here, the existence distribution is illustrated with only two levels of existence probabilities in the central part and the peripheral part. However, by setting the existence probabilities of three or more levels, the closer to the central part, the more characteristic The existence probability of the point may be high, and the existence probability may be lowered stepwise as the distance from the center portion increases.

  When the probability distributions 32a to 32d of (d) are obtained, the feature point at time T2 is detected as follows from this and the captured image of (b). First, cumulative correlation values are calculated when the captured image of (b) and the presence distributions 32a to 32d of (d) are superimposed while being shifted one pixel at a time. At this time, an edge intersection point portion of the captured image is set to 1 and other portions are set to 0, and a value obtained by multiplying this by the existence probability set in the existence distribution is accumulated for each pixel, thereby accumulating the correlation. A value is calculated. Next, the superposition state (superposition position) of the captured image and the existence distribution when the cumulative correlation value calculated in this way becomes maximum is obtained, and the point where the edge intersection and the existence distribution overlap in that state is obtained. Ask. This point is detected as a feature point. Such a feature point detection method is called feature point shape pattern matching.

  When a feature point at time T2 is detected by the above feature point shape pattern matching, a vehicle region is set for the feature point. As a result, as shown in (e), the edge intersections 22a to 22d are detected as feature points, and the vehicle region 12 is set. As described above, the vehicle tracking process when the moving speed of the edge intersection is slow is performed.

  The feature points 22a to 22d and the vehicle area 12 are registered as new vehicle patterns. Thereby, the vehicle pattern of (c) is updated. By updating the vehicle pattern in this manner, even if the other vehicle 100 approaches or leaves the host vehicle, the area occupied in the captured image is enlarged or reduced with respect to the captured image. The other vehicle 100 can be extracted from the captured image using the vehicle pattern.

  Next, the vehicle tracking process when the moving speed of the edge intersection shown in FIG. 3 is fast will be described. FIGS. 3A and 3B respectively show captured images acquired by the camera 1 at time T1 and time T2 of the next image frame. In these captured images, the other vehicle 100 approaching from behind the host vehicle and the object 200 stationary on the side of the road are captured. From the captured image of (a), the edge intersections 23a, 23b, 23c and 23d of the other vehicle 100 and the edge intersections 25a, 25b and 25c of the object 200 are extracted. Similarly, edge intersections 24a, 24b, 24c, 24d and 24e of the other vehicle 100 and edge intersections 26a, 26b and 26c of the object 200 are extracted from the captured image of (b). In addition, as for each of these edge intersections, what represents the same point in a vehicle or an object is mutually matched like the case of FIG. For example, the edge intersection 23a of (a) representing the same point of the other vehicle 100 and the edge intersection 24a of (b) are associated, and the edge intersection 25a of (a) representing the same point of the object 200 and the edge of (b) The intersection point 26a is associated.

  As described with reference to FIG. 2, the moving speed in the captured image of each edge intersection at time T2 is obtained from the amount of change in position between the associated edge intersections, and the average value of the moving speeds for the edge intersections of the same vehicle. Is calculated. That is, the average value of the moving speed is calculated for the edge intersections 24a to 24d of the other vehicle 100. Here, it is assumed that the average value of the moving speeds of the edge intersections 24a to 24d is larger than a predetermined value. Note that the edge intersections 26a to 26c of the object 200 are determined not to represent the vehicle from the moving speed, and the processes described below are not performed.

  Here, in the captured image of (a), it is assumed that the edge intersections 23a to 23d of the other vehicle 100 among the edge intersections described above are detected as feature points of the other vehicle 100 at time T1. In addition, it is assumed that the vehicle region 13 is set for the feature points 23a to 23d. At time T1, the feature points 23a to 23d and the vehicle area 13 are registered as vehicle patterns of the other vehicle 100 as shown in (c). Each of the feature points 23a to 23d has a velocity vector as indicated by an arrow in the figure, and moves in a direction and a speed determined by the velocity vector.

  As described above, when the average value of the moving speed of the edge intersection at time T2 is larger than the predetermined value, the feature point detection vector 50 and the vehicle pattern detection area 51 shown in (d) are based on the vehicle pattern of (c). The vehicle pattern speed is calculated. The direction and the magnitude of the feature point detection vector 50 are obtained from the average values of the directions and magnitudes of the velocity vectors respectively possessed by the feature points 23a to 23d. The vehicle pattern detection area 51 is set by obtaining the range of the predicted position of the feature point at time T2 from the movement history of the feature points 23a to 23d. The position of the vehicle pattern detection area 51 set here is fixed with respect to the captured image, unlike the probability distributions 32a to 32d in FIG.

  When the feature point detection vector 50 and the vehicle pattern detection area 51 of (d) are obtained, the feature point at time T2 is detected from this and the captured image of (b). At this time, among the edge intersections extracted in the captured image of (b), an edge that exists in the vehicle pattern detection region 51 and that has a velocity vector difference with respect to the feature point detection vector 50 within a predetermined value set in advance. Intersection points are detected as feature points. Such a feature point detection method is called feature point speed pattern matching.

  When a feature point at time T2 is detected by the above feature point speed pattern matching, a vehicle region is set for the feature point. As a result, as shown in (e), the edge intersections 24a to 24e are detected as feature points, and the vehicle region 14 is set. As described above, the vehicle tracking process when the moving speed of the edge intersection is fast is performed.

  The feature points 24a to 24e and the vehicle area 14 are registered as new vehicle patterns. Thereby, the vehicle pattern of (c) is updated. By updating the vehicle pattern in this way, even when the position and number of edge intersections change due to the change in the appearance of the other vehicle 100 in the captured image, the feature points are added or deleted accordingly. The other vehicle 100 can be extracted from the captured image using the pattern.

  A flowchart of the processing described above is shown in FIG. This flowchart is executed for each image frame for each frame rate. In step S10, edge intersection extraction processing shown in the flowchart of FIG. 5 described later is performed, and edge intersections are extracted from the captured image. In step S20, the new vehicle detection process as described above is performed by executing the flowchart of FIG. 6 described later on the edge intersection extracted in step S10. In step S30, among the edge intersections extracted in step S10, the edge intersections that have already been detected as feature points in the previous processing are executed by executing the flowchart of FIG. Vehicle tracking process. After step S30 is executed, the flowchart of FIG. 4 is terminated.

  The contents of the edge intersection extraction process executed in step S10 will be described with reference to the flowchart of FIG. In step S100, an image captured by the camera 1 is acquired by reading an image stored in the image memory 2. In step S101, a vertical edge and a horizontal edge are extracted from the captured image acquired in step S100. In the next step S102, an intersection of the vertical edge and the horizontal edge extracted in step S101 is extracted. In step S103, the moving speed on the screen of the edge intersection extracted in step S102 is calculated. As described above, this moving speed can be calculated by obtaining the position change amount from the edge intersection in the previous captured image. If step S103 is performed, step S10 of FIG. 4 will be complete | finished. In this way, edge intersection extraction processing is performed.

  Next, the contents of the new vehicle detection process executed in step S20 will be described with reference to the flowchart of FIG. In step S200, it is determined whether or not the Y coordinate value of each edge intersection extracted by the edge intersection extraction process is equal to or less than a predetermined value. If the Y coordinate value is less than or equal to the predetermined value, the process proceeds to step S201, and if greater than the predetermined value, the process proceeds to step S204. Here, the position of the edge intersection on the screen in the vertical direction is represented by a Y coordinate value, and this Y coordinate value is defined to increase from the upper end to the lower end of the screen. That is, as the Y coordinate value is smaller, the edge intersection is located above the captured image on the screen, and conversely, as the Y coordinate value is larger, the edge intersection is located below the captured image on the screen. .

  When the process proceeds from step S200 to step S201, the feature point is detected by the pattern matching process described above. In step S201, a basic vehicle pattern stored in advance is read. In the next step S202, feature points are detected by performing pattern matching processing using this basic vehicle pattern. In step S203, a vehicle area is set for the feature point detected in step S202, and the arrangement of the feature point and the vehicle area are registered as a vehicle pattern. After the feature points are detected using the pattern matching process as described above, step S20 in FIG. 4 is terminated. In this way, the new vehicle detection process is performed on the edge intersection located above in the captured image.

  On the other hand, when the process proceeds from step S200 to step S204, the feature points are detected by the grouping process described above. In step S204, feature points are detected by performing grouping processing. In step S205, for the feature points detected in step S204, the average value of the speed and direction of the feature points in each group represents a vehicle in which the feature point group is traveling in the direction of the host vehicle. It is determined whether or not there is. At this time, the feature points included in the group determined not to represent the vehicle are excluded from the subsequent processing. By the processing in step S205, feature points detected from the edge intersections of the stationary object are excluded. In step S206, a vehicle area is set for the feature points not excluded in step S205, and the arrangement of the feature points and the vehicle area are registered as a vehicle pattern. After the feature points are detected using the grouping process as described above, step S20 in FIG. 4 is terminated. In this way, the new vehicle detection process is performed on the edge intersection located below in the captured image.

  Next, the contents of the vehicle tracking process executed in step S30 will be described with reference to the flowchart of FIG. In step S300, it is determined whether or not the moving speed of the edge intersection is below a predetermined value. Note that the edge intersections to be determined here are edge intersections that have already been detected as feature points in the previous processing as described above. If the moving speed of the edge intersection is less than or equal to a predetermined value, the process proceeds to step S301, and if greater than the predetermined value, the process proceeds to step S304.

  When the process proceeds from step S300 to step S301, the feature points are detected by the feature point shape pattern matching process as described with reference to FIG. In step S301, based on the vehicle pattern registered in step S203 or S206 of FIG. 6 or step S303 or S306 of the previous process, vehicle patterns exist as shown in probability distributions 32a to 32d shown in FIG. Calculate the distribution. In step S302, feature points are detected by performing feature point shape pattern matching using the vehicle pattern presence distribution calculated in step S301. In step S303, the vehicle pattern is updated by setting a vehicle area for the feature point detected in step S302 and registering the arrangement of the feature point and the vehicle area as a vehicle pattern. After the feature points are detected using the feature point shape pattern matching process as described above, step S30 in FIG. 4 is terminated. In this way, the vehicle tracking process is performed on the edge intersection where the moving speed is slow.

  On the other hand, when the process proceeds from step S300 to step S304, the feature point is detected by the feature point speed pattern matching process as described with reference to FIG. In step S304, based on the vehicle pattern registered in step S203 or S206 of FIG. 6 or step S303 or S306 of the previous process, the feature point detection vector 50 and the vehicle pattern detection area 51 shown in FIG. The vehicle pattern speed is calculated as follows. In step S305, feature points are detected by performing feature point speed pattern matching using the vehicle pattern speed calculated in step S304. In step S306, a vehicle area is set for the feature point detected in step S305, and the vehicle pattern is updated by registering the arrangement of the feature point and the vehicle area as a vehicle pattern. After the feature points are detected using the feature point speed pattern matching process as described above, step S30 in FIG. 4 is terminated. In this way, the vehicle tracking process is performed on the edge intersection where the moving speed is fast.

  By performing the processing as described above for each image frame of the captured image, the vehicle area of the other vehicle is detected. Based on the position of the vehicle area thus detected on the screen, the distance in real space of the other vehicle to the host vehicle is calculated. Moreover, the speed in real space with respect to the own vehicle of the other vehicle is calculated based on the positional change amount of the vehicle region between the captured images acquired at different times. When the calculated distance or speed satisfies a predetermined condition, a marker for indicating the presence of an obstacle is displayed on the monitor 4. For example, a marker is displayed for other vehicles whose distance is equal to or less than a predetermined value and whose speed is equal to or greater than a predetermined value. In this way, the driver of the host vehicle is notified of the presence of another vehicle traveling behind the host vehicle.

According to the embodiment described above, the following operational effects can be obtained.
(1) A captured image behind the host vehicle is acquired by the camera 1, a vehicle detection method is further selected, and the other vehicle 100 behind the host vehicle is detected based on the captured image using the selected vehicle detection method. It was decided to. Since it did in this way, the suitable vehicle detection method can be selected according to a condition, and the stable vehicle detection can be performed.

(2) In the new vehicle detection process, since one of the pattern matching process (steps S201 to 203) and the grouping process (steps S204 to 206) is selected as the vehicle detection method, a new captured image is displayed. The captured other vehicle can be reliably detected.

(3) The vehicle detection method is selected based on the position on the screen in the captured image of the edge intersection. Specifically, when the edge intersection is located above a predetermined position on the screen, pattern matching processing is performed. If the position is below the predetermined position, the grouping process is selected (step S200). Since it did in this way, according to the distance in the real space of the other vehicle with respect to the own vehicle, an appropriate vehicle detection method can be selected also when the other vehicle's appearance differs.

(4) In the vehicle tracking process, as the vehicle detection method, one of the feature point shape pattern matching process (steps S301 to 303) and the feature point speed pattern matching process (steps S304 to 306) is selected and performed. Even when another vehicle that has already been captured in the captured image is tracked and detected, it can be reliably detected.

(5) The vehicle detection method is selected based on the moving speed in the captured image of the edge intersection. Specifically, when the average value of the moving speed of the edge intersection representing the same vehicle is equal to or less than a predetermined value, the feature point The shape pattern matching process is selected, and if the average value is larger than the predetermined value, the feature point speed pattern matching process is selected (step S300). Since it did in this way, according to the moving speed in the real space of the other vehicle with respect to the own vehicle, when a mode that the appearance of the other vehicle changes changes, an appropriate vehicle detection method can be selected.

  In the above-described embodiment, an example has been described in which the rear side of the host vehicle is imaged using the camera 1 and another vehicle approaching from the rear side of the host vehicle is detected based on the captured image. However, the present invention is not limited to this content, and when other directions around the host vehicle, for example, the front and sides are imaged, and other vehicles existing in that direction are detected based on the captured image. Is also applicable.

  In the above embodiment, the imaging means is realized by the camera 1, and the other means are realized by the processing of the microcomputer 3. However, this is merely an example, and unless the characteristics of the present invention are impaired, The components are not limited to the above embodiment.

It is a figure which shows the structure of the vehicle detection apparatus by one Embodiment of this invention. It is a figure for demonstrating the content of the vehicle tracking process in case the moving speed of an edge intersection is slow. It is a figure for demonstrating the content of the vehicle tracking process in case the moving speed of an edge intersection is quick. It is a flowchart of the process performed when detecting another vehicle. It is a flowchart which shows the content of edge intersection extraction processing. It is a flowchart which shows the content of the new vehicle detection process. It is a flowchart which shows the content of a vehicle tracking process.

Explanation of symbols

1: Camera 2: Image memory 3: Microcomputer 4: Monitor 100: Other vehicle

Claims (4)

Imaging means for acquiring a captured image around the host vehicle;
Candidate point extraction means for extracting candidate points as candidate feature points representing a part of another vehicle existing around the host vehicle in the captured image from the captured image;
Selection means for selecting a vehicle detection method for detecting the other vehicle based on the position or moving speed of the candidate point extracted by the candidate point extraction means ;
A vehicle detection apparatus comprising: vehicle detection means for detecting the other vehicle using a vehicle detection method selected by the selection means based on the captured image . The vehicle detection device according to claim 1 ,
The selection means selects a pattern matching process as the vehicle detection method when the candidate point is positioned above a predetermined position on the screen, and the candidate point is positioned below the predetermined position on the screen. If so, select the grouping process as the vehicle detection method,
The vehicle detection means includes
(1) When a pattern matching process is selected by the selection unit, based on a comparison result between an image pattern within a predetermined range around the candidate point and an image pattern stored in advance, the candidate point is selected from the candidate points. Detecting the feature points;
Detecting the other vehicle by setting a vehicle area corresponding to the other vehicle in the captured image based on the position of the detected feature point;
(2) When the grouping process is selected by the selection unit, the candidate points are grouped with each other among the candidate points that have a velocity vector difference in the captured image that is equal to or less than a predetermined value. Detecting the feature points from
A vehicle detection device that detects the other vehicle by setting the vehicle region in the captured image based on the position of the detected feature point. In the vehicle detection device according to claim 1 or 2 ,
When the average value of the moving speeds of candidate points representing the same vehicle is equal to or less than a predetermined value, the selection unit selects a feature point shape pattern matching process as the vehicle detection method, and the average value is less than the predetermined value. If larger, select the feature point speed pattern matching process as the vehicle detection method,
The vehicle detection means includes
(1) When the feature point shape pattern matching process is selected by the selection unit, based on the relative positions and movement speeds of the plurality of feature points detected in the past, Set the relative existence probability of each other,
Detecting the feature point from the candidate points based on a set existence probability;
Detecting the other vehicle by setting a vehicle area corresponding to the other vehicle in the captured image based on the position of the detected feature point;
(2) When the feature point speed pattern matching process is selected by the selection unit, based on each average value of the direction and the size of the speed vector in the captured image respectively possessed by a plurality of feature points detected in the past , Calculate feature point detection vector,
Based on the current expected position range of the plurality of feature points, a vehicle pattern detection area is set in the captured image,
The feature point is detected from the candidate points by obtaining a candidate point that exists in the vehicle pattern detection region and that has a velocity vector difference with respect to the feature point detection vector within a predetermined value. ,
A vehicle detection apparatus for detecting the other vehicle by setting a vehicle area corresponding to the other vehicle in the captured image based on the position of the detected feature point. In the vehicle detection device according to any one of claims 1 to 3,
When the selection unit newly detects another vehicle captured in the captured image, the selection unit selects either the pattern matching process or the grouping process as the vehicle detection method based on the position of the candidate point. When tracking other vehicles already captured in the captured image, either the feature point shape pattern matching process or the feature point speed pattern matching process is used as the vehicle detection method based on the moving speed of the candidate point. A vehicle detection device characterized by selecting the above.

Images