JP2005149250A - Vehicle detection method and vehicle detection system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To simply and quickly recognize and detect a vehicle ahead by extracting a vehicle feature quantity stably through abstraction by a small scale robust structure resistant to noise and the like. <P>SOLUTION: A rectangular prediction area enclosing a vehicle ahead of an associated vehicle is set as a vehicle candidate area in a pickup image from a monocular camera mounted on the associated vehicle. Processing by a robust voting method resistant to noise and the like is applied to a vertical edge image and a horizontal edge image in the area to form a plurality of vertical edge side voting planes and horizontal edge side voting planes with different voting positions. From the coordinates of a peak value on the composite voting plane into which the voting results are superposed, the vehicle center position at the center of the vehicle ahead of the associated vehicle in relation to the vehicle width and vehicle height is stably extracted as a vehicle feature quantity. The vehicle ahead of the associated vehicle is recognized and detected on the basis of the vehicle feature quantity. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a vehicle detection method and a vehicle detection device for recognizing and detecting a vehicle ahead of a running vehicle from a captured image in front of the vehicle of an image sensor mounted on the vehicle.

  Conventionally, a vehicle such as an advanced safety vehicle called ASV is equipped with an image sensor, and recognizes a vehicle in front of the host vehicle (hereinafter referred to as a preceding vehicle) by image processing of a captured image of the image sensor. The rear-end collision is determined based on the detection.

  The image processing is image template matching pattern recognition processing. By this recognition processing, the contour line of the preceding vehicle is extracted, the vehicle pattern of the preceding vehicle determined by the extracted contour line, and various prototypes stored in advance. The vehicle pattern is discriminated from the vehicle pattern, and the vehicle ahead is recognized and detected from the vehicle pattern as the vehicle feature amount (see, for example, Patent Document 1).

JP 7-182484 A (paragraph number [0011], FIG. 9)

  When pattern recognition is performed by extracting the contour line of the preceding vehicle as in the prior art, the extraction result is easily affected by noise (disturbance) such as changes in the brightness of the driving environment. It is necessary to change the threshold value dynamically according to changes in the brightness of the driving environment, etc., but this threshold value changes the vehicle pattern based on the extracted contour line even for the same vehicle. In some cases, the vehicle pattern is significantly different from the corresponding prototype vehicle pattern, making it difficult to recognize the preceding vehicle, which may lead to an undetectable situation.

  Further, it is practically impossible to prepare prototype vehicle patterns for all vehicle types as vehicle feature values, and there is a problem that recognition accuracy cannot be improved.

  In addition, since it is necessary to perform time-consuming complicated pattern recognition processing based on a large number of prototype vehicle patterns, there is also a problem that it is not possible to easily and quickly recognize and detect the preceding vehicle with a small configuration. is there.

  And in order to solve these problems, it is desirable to recognize the vehicle ahead by abstracting (simplifying) the vehicle feature amount with a robust and robust configuration resistant to noise and the like. No invention has been invented.

  The present invention is a vehicle detection method and a vehicle detection method having a specific configuration in which vehicle feature values are abstracted and stably extracted by a robust and small-sized configuration that is strong against noise and the like, and a front vehicle is recognized and detected easily and quickly. An object is to provide an apparatus.

  In order to achieve the above-described object, the vehicle detection method of the present invention sets a rectangular prediction region surrounding a preceding vehicle as a vehicle candidate region in a captured image in front of the own vehicle of an image sensor mounted on the own vehicle, For the vertical edge image of the vehicle candidate area, vote +1 to the position where the vertical edge of the left half of the vehicle candidate area is moved to the right of the set length T *, and set each vertical edge of the right half of the vehicle candidate area to the set length Voting +1 at the position moved to the left side of T * to form a vertical edge side voting plane, the set length T * is Tmin ≦ T * ≦ Tmax (Tmin and Tmax are the set minimum length and maximum length). And +1 is voted at the position where the horizontal edge of the lower half of the vehicle candidate area is moved upward by the set length H * with respect to the horizontal edge image of the vehicle candidate area. +1 is voted at the position where each horizontal edge of the minute is moved to the lower side of the set length T * to form a horizontal edge side voting plane, and the set length H * is set to Hmin ≦ H * ≦ Hmax, (Hmin, Hmax is repeated in a variable range within the set minimum length and maximum length), and the vehicle width direction and vehicle height of the preceding vehicle are determined from the coordinates of the peak value of the composite voting plane obtained by superimposing the voting results of the two edge side voting planes. A vehicle center position in the center of the direction is extracted as a vehicle feature value, and the vehicle ahead is recognized and detected from the vehicle feature value (Claim 1).

  In the vehicle detection method of the present invention, the rectangular prediction area surrounding the preceding vehicle is an area where vertical and horizontal edges of the captured image in front of the own vehicle are concentrated (claim 2). A distance measurement sensor such as a laser radar or a millimeter wave radar is mounted on the vehicle, and a rectangular prediction region surrounding the vehicle ahead of the host vehicle is set from the position of the distance measurement target of the distance measurement sensor. ).

  Furthermore, the vehicle detection method of the present invention is characterized in that the image sensor is a monocular camera.

  Next, the vehicle detection device of the present invention is an image sensor that is mounted on the host vehicle and captures the front of the host vehicle, and a candidate that sets, as a vehicle candidate region, a rectangular prediction region that surrounds the front vehicle in the captured image of the image sensor. For the area setting means and the vertical edge image of the vehicle candidate area, vote +1 to the position where the vertical edge of the left half of the vehicle candidate area is moved to the right of the set length T *, and each of the right half of the vehicle candidate area The vertical edge side voting plane is formed by voting +1 to the position where the vertical edge is moved to the left side of the set length T *, and the set length T * is set to Tmin ≦ T * ≦ Tmax (Tmin and Tmax are set) Vertical edge side voting processing means that repeats in a variable range within a range of (minimum length, maximum length) and a horizontal edge image of the vehicle candidate area, the horizontal edge of the lower half of the vehicle candidate area is set to a set length H * above Voting +1 to the moved position, voting +1 to the position where each horizontal edge of the upper half of the vehicle candidate area has moved to the lower side of the set length T * to form a horizontal edge side voting plane, Horizontal edge side voting processing means that repeats the set length H * in a range of Hmin ≦ H * ≦ Hmax (where Hmin and Hmax are set minimum length and maximum length), and the voting results of the both edge side voting planes Voting result combining means for forming a combined voting plane in which the vehicle is overlapped, and the vehicle center position in the center in the vehicle width direction and the vehicle height direction of the vehicle ahead of the host vehicle from the coordinates of the peak value of the combined voting plane Vehicle feature quantity extraction means for extracting the vehicle in front of the vehicle, and detecting and detecting the preceding vehicle from the vehicle feature quantity.

  The vehicle detection apparatus of the present invention is characterized in that the candidate region setting means sets the rectangular prediction region surrounding the preceding vehicle as a region where vertical and horizontal edges of the captured image in front of the host vehicle are concentrated (claim). Item 6) A range sensor such as a laser radar is mounted on the host vehicle, and a rectangular prediction region surrounding the vehicle ahead of the host vehicle is set by the candidate region setting means from the position of the target of range measurement of the range sensor. (7).

  Furthermore, the vehicle detection device of the present invention is characterized in that the image sensor is a monocular camera.

  First, according to the configurations of claims 1 and 5, since the vehicle ahead of the captured image of the image sensor can be regarded as a rectangle, a rectangular prediction region is set as a vehicle candidate region in the captured image, and the vertical edge of this vehicle candidate region Obtain images and horizontal edge images.

  At this time, the vertical edge and horizontal edge of the vehicle ahead in the vehicle candidate area are almost symmetrical with respect to the center position of the vehicle width and vehicle height of the vehicle ahead, left and right ends, upper and lower ends of the vehicle. Occurs in the department.

  Therefore, while changing the set length T * in the range of Tmin ≦ T * ≦ Tmax, +1 is voted at the position where the vertical edge of the left half of the vehicle candidate area is moved to the right side of the set length T *, and the right of the vehicle candidate area is If voting +1 is formed at the position where each half vertical edge is moved to the left side of the set length T * to form each vertical edge side voting plane, the set length T * is half of the vehicle width of the preceding vehicle on the image. In the vertical edge side voting plane, the voting position of the left half edge of the forward vehicle and the voting position of the right half edge coincide at the center position in the vehicle width direction of the forward vehicle, and the voting value at that position is +2. .

  In addition, while changing the set length H * within the range of Hmin ≦ H * ≦ Hmax, +1 is voted at the position where the horizontal edge of the lower half of the vehicle candidate area is moved above the set length H *, When voting +1 at the position where each half horizontal edge is moved to the lower side of the set length H * to form each horizontal edge side voting plane, the set length H * is half of the vehicle height of the preceding vehicle on the image. In the horizontal edge side voting plane, the voting position of the lower half edge coincides with the voting position of the upper half edge at the center position in the vehicle height direction of the preceding vehicle, and the voting value at that position becomes +2.

  Therefore, in the composite voting plane obtained by superimposing the voting results of these voting planes, the voting value at the center of the vehicle center in the vehicle width direction and the vehicle height direction is a peak value of +4. From the above, the vehicle center position in the center in the vehicle width direction and the vehicle height direction can be stably extracted as an abstract vehicle characteristic amount of the preceding vehicle, and the forward vehicle is recognized based on the vehicle characteristic amount. Can be detected.

  In this case, it is not necessary to store a large number of prototype edge pattern data and the like, and instead of complicated processing that takes time for pattern recognition, simple processing of a voting method resistant to noise or the like may be performed. The vehicle features are abstracted and stably extracted with a robust, small-scale and inexpensive configuration that is resistant to noise, etc., and the vehicle ahead can be easily recognized and detected from the extracted vehicle features. Can be provided.

  According to the second and sixth aspects of the present invention, since vertical and horizontal edges are concentrated in the area of the front vehicle in the captured image in front of the host vehicle, the vertical and horizontal of the captured image in front of the host vehicle are generated. By setting the region where the edges of the vehicle are concentrated as a prediction region and setting this region as a vehicle candidate region, the vehicle candidate region is set to an appropriate region including the preceding vehicle only by image processing of the captured image of the image sensor. The vehicle ahead can be detected reliably.

  Furthermore, according to the configurations of claims 3 and 7, the distance measurement position of the distance measurement sensor mounted on the host vehicle is often the position of a reflector or the like of the vehicle ahead, and the prediction area is surrounded by these positions. By setting the vehicle candidate area, the vehicle candidate area can be set to an appropriate area including the preceding vehicle, and the preceding vehicle can be reliably detected.

  Next, according to the configurations of claims 4 and 8, the image sensor can be formed by a monocular camera that is less expensive than a stereo camera, and the vehicle ahead can be accurately recognized and detected with a more inexpensive and simple configuration. it can.

  Next, in order to describe the present invention in more detail, an embodiment thereof will be described in detail with reference to FIGS.

  1 is a block diagram of the vehicle detection device, FIG. 2 is a flowchart for explaining the operation of FIG. 1, FIG. 3 is an actual measurement diagram for explaining the operation of FIG. 1, and FIG. 4 is a voting on the vertical edge side of FIG. FIG. 5 is an explanatory diagram of the voting process on the horizontal edge side of FIG. 1, and FIG. 6 is an explanatory diagram of the composite voting plane of FIG.

<Configuration>
First, the configuration of the vehicle detection device in FIG. 1 will be described.

  As shown in FIG. 1, this apparatus includes a monocular camera 2 having a two-dimensional solid-state imaging device (CCD) configuration as an image sensor in front of a host vehicle (vehicle) 1 such as an advanced safety vehicle (ASV). .

  This monocular camera 2 repeatedly shoots in front of the host vehicle and sends, for example, digital data of a monochrome captured image to the ECU 3 for recognition processing, which is composed of a microcomputer, in real time.

  Further, the ECU 3 and the memory 4 form a recognition processing unit 5 for recognizing and detecting the vehicle ahead, and the microcomputer executes the preset recognition processing program of steps S1 to S8 of vehicle detection in FIG. Accordingly, the recognition processing unit 5 includes the following units (a) to (h).

(A) Candidate area setting means This means sets a rectangular prediction area surrounding the preceding vehicle in the captured image of the image sensor 2 as a vehicle candidate area.

  In this area setting, the vertical and horizontal edges of the image are concentrated on the vehicle part, so the edge distribution of the captured image is detected and the area on the image where the vertical and horizontal edges are concentrated is detected. Then, a rectangular area having a set size including all or part of the detection area is set as a prediction area, and this prediction area is set as a vehicle candidate area.

(B) Vertical edge image forming means This means detects the vertical edge of the vehicle candidate area from the luminance change of the vehicle candidate area of the photographed image, and forms a vertical edge image of the detected edge.

(C) Horizontal edge image forming means This means detects the horizontal edge of the vehicle candidate area from the luminance change of the vehicle candidate area of the photographed image, and forms a horizontal edge image of the detected edge.

(D) Vertical edge side voting processing means This means performs voting by voting on the vertical edge image of the vehicle candidate area.

  In this vote, +1 is voted at the position where the left half vertical edge of the vehicle candidate area is moved to the set length T * right side, and each vertical edge of the right half of the vehicle candidate area is moved to the position moved to the left side of the set length T *. By voting +1, a two-dimensional vertical edge side voting plane is formed with the horizontal axis direction being the vehicle width direction (left-right direction of the image) and the vertical axis direction being the vehicle height direction (up-down direction of the image). is there.

  Then, in order to detect the center position in the vehicle width direction of the preceding vehicle from the result of this voting, the vertical edge side voting processing means sets the set length T * to Tmin ≦ T * ≦ Tmax (Tmin and Tmax are set) In the range of (minimum length, maximum length), for example, the above-mentioned voting is repeated while changing by a set unit length, and a plurality of vertical edge side voting planes are formed by changing the set length T * little by little.

(E) Horizontal edge side voting processing means This means performs a voting voting on the horizontal edge image of the vehicle candidate area.

  In this voting, +1 is voted at the position where the lower horizontal edge of the vehicle candidate area is moved upward by the set length H *, and the position where the upper horizontal edge of the vehicle candidate area is moved downward by the set length H *. By voting +1, a horizontal edge voting plane is formed in which the horizontal axis direction and the vertical axis direction are the same as the vertical edge voting plane.

  Then, in order to detect the center position in the vehicle height direction of the preceding vehicle from the result of this voting, the horizontal edge side voting processing means sets the set length H * to Hmin ≦ H * ≦ Hmax (Hmin and Hmax are set). In the range of (minimum length, maximum length), for example, the above-mentioned voting is repeated while changing by a set unit length, and a plurality of horizontal edge side voting planes are formed by changing the set length H * little by little.

  In each voting plane on the vertical edge side and the horizontal edge side, actually, voting points are set in a matrix at intervals of unit lengths set in the horizontal axis direction and the vertical axis direction. Alternatively, voting based on the presence or absence of a horizontal edge is performed.

(F) Voting Result Combining Means This means superimposes each vertical edge side voting plane and each horizontal edge side voting plane to form a composite voting plane obtained by superimposing the voting results of both edge side voting planes.

(G) Vehicle feature amount extraction means This means is a vehicle center position in the vehicle width direction and in the vehicle height direction that is a position half the vehicle width and vehicle height of the preceding vehicle from the coordinates of the peak value of the composite voting plane. Are extracted as vehicle feature values.

(H) Vehicle Recognizing Means This means recognizes and detects the forward vehicle by grasping the presence and position of the forward vehicle from the vehicle center position as the vehicle feature amount.

  Then, the result of this detection is sent from the ECU 4 of the recognition processing unit 6 to the ECU of the collision determination process for determining the possibility of collision of the host vehicle 1, and this ECU determines the possibility of collision with the preceding vehicle or the like. Then, control of traveling and steering necessary for collision avoidance is performed.

<Operation>
Next, the operation of the apparatus of FIG. 1 based on the above configuration will be described with reference to the flowchart of FIG.

  First, while the host vehicle 1 is traveling, the laserator 2 repeatedly shoots the front of the host vehicle 1, and when a shot image such as the shot image Pi of FIG. 2, for example, the two-dimensional filter processing is performed on the photographed image Pi to detect the edge of the photographed image Pi, and the width of the region where the vertical and horizontal edges of the image Pi are concentrated or Based on the height and the ratio of the set vehicle width / vehicle height, a rectangular area approximately the size of one vehicle based on the width or height of this area is set as a prediction area, and this prediction area is For example, it is set in the vehicle candidate area Q in FIG.

  Next, in steps S2 and S3 in FIG. 2, the vertical edge image forming unit and the horizontal edge image forming unit operate to form, for example, the vertical edge image Pa and the horizontal edge image Pb in FIG.

  Further, in steps S4 and S5 of FIG. 2, the vertical edge side voting processing means and the horizontal edge side voting processing means operate to perform voting, and the voting plane (vertical edge side voting plane) and horizontal edge for the vertical edge image Pa are obtained. A voting plane (horizontal edge side voting plane) for the image Pb is formed.

  By the way, as represented by the Hough transform, it has been confirmed in many experiments that the voting algorithm is robust against noise and robust.

  Therefore, the present invention utilizes the fact that the vehicle can be approximated as “approximately rectangular (quadrangle) with approximately left-right symmetry by a robust voting method that is resistant to noise, and as an optimal feature amount of the preceding vehicle, the vehicle width direction and The vehicle center position in the center in the vehicle height direction is accurately detected, and the vehicle ahead is recognized and detected based on this detection.

  Therefore, the vertical edge side voting processing unit and the horizontal edge voting processing unit perform voting processing as will be described below to form a vertical edge side voting plane and a horizontal edge side voting plane.

  First, in order to simplify the description, the vertical edge image Pa of the front vehicle is, for example, as shown in FIG. 4, the positions of the left and right end portions of the front vehicle, that is, the T-axis in the vehicle width (lateral width) direction. For example, as shown in FIG. 5, the horizontal edge image Pb of the front vehicle is the position of the upper and lower ends of the front vehicle, that is, the vehicle height (height) direction. It is assumed that the image is an image of the edge Gb at positions Hu and Hd in the H-axis direction in FIG.

  In this case, when the vehicle width on the image of the preceding vehicle is t and the vehicle height is h, the positions Tl and Tr of the two vertical edges Ga are relative to the center position Tc in the T-axis direction of the preceding vehicle. The positions Hu and Hd of the two horizontal edges Gb are half of the vehicle height h with respect to the center position Hc in the H-axis direction of the preceding vehicle. The symmetry position is h / 2 apart.

  On the other hand, the minimum value Tmin and the maximum value Tmax of the set length T * are, for example, vertical edges so that voting is performed at the center positions Tc and Hc regardless of the vehicle width t and vehicle height h of the preceding vehicle. The horizontal width of the image Pa (the horizontal width of the vehicle candidate area Q) is set to 1/4 and 3/4. Similarly, the minimum value Hmin and the maximum value Hmax of the set length H * are, for example, the horizontal width ( The lateral width of the vehicle candidate area Q) is set to 1/4 and 3/4.

  Then, the vertical edge side voting processing means adds +1 to the position Tl + T * obtained by moving the left vertical edge Ga of the vertical edge image Pa to the set length T * to the right based on the set length T * (Tmin ≦ T * ≦ Tmax). Voting and voting +1 to the position Tr-T * where the vertical edge Ga on the right side of the vertical edge image Pt has been moved to the set length T * left side to form the vertical edge side voting plane Ra of FIG. T * is repeated while varying the unit length by the minimum length Tmin to the maximum length Tmax or vice versa.

  In this repetition, when the set length T * = t / 2, as shown in the voting plane Ra of T * = t / 2 in FIG. The plane Ra is a plane having a voting value of +2 at the position Tc.

  Similarly, the horizontal edge side voting processing means is based on the set length H * (Hmin ≦ H * ≦ Hmax) at the position Hd + H * where the vertical edge Ga on the lower side of the horizontal edge image Pb has been moved to the upper side of the set length H *. +1 is voted, and +1 is voted at the position Hu-H * where the upper horizontal edge Gb of the horizontal edge image Pb is moved to the lower side by the set length H * to form the horizontal edge side voting plane Rb of FIG. The set length H * is repeated while changing the unit length by the minimum length Hmin to the maximum length Hmax or vice versa.

  In this repetition, when the set length H * = h / 2, as shown in the voting plane Rb of H * = h / 2 in FIG. 5, the voting positions of the horizontal edges Gb on both sides become the center position Hc. The voting plane Rb at that time is a plane in which the voting value at the position Hc is +2.

  Next, in step S6 of FIG. 2, the voting result synthesizing means superimposes the voting result of each vertical edge side voting plane and the voting result of each horizontal edge side voting plane, for example, the synthetic voting plane Rab shown in FIG. Form.

  At this time, the voting result (+2) of the vertical edge side voting plane Ra having the set length T * = t / 2 and the voting result (+2) of the horizontal edge side voting plane Rb having the set length H * = h / 2 are overlapped. As a result, the voting result at the vehicle center position at the coordinates (T, H) = (Tc, Hc) of the combined voting plane Rab becomes a peak value (maximum value) of +4.

  Actually, the voting position of the peak value does not always coincide with the vehicle center position (Tc, Hc), but the set values T * and H * are set to Tmin ≦ T * ≦ Tmax and Hmin ≦ H * ≦ Hmax. If a plurality of voting planes with slightly different voting positions are prepared and overlapped, the peak value is always in a very small range centered on the coordinates (Tc, Hc) of the peak value by the so-called convolution effect. Appears, and the vehicle center position (Tc, Hc) can be accurately detected from the coordinates.

  Then, when the voting value of each voting point on the combined voting plane Rab is expressed by the luminance proportional to the voting value, the combined actual voting result becomes, for example, as shown in the combined voting plane Rab in FIG. Also in the voting plane Rab of FIG. 3, the peak value appears in a very small range centered on the coordinate (Tc, Hc) of the peak value at the center.

  Next, in step S7 in FIG. 2, the vehicle feature amount extraction unit operates to extract the peak value coordinates (Tc, Hc) as the vehicle feature amount of the preceding vehicle.

  Then, based on the vehicle feature amount and the distance measured between the own vehicle 1 and the preceding vehicle, the actual vehicle center position of the preceding vehicle is grasped, the preceding vehicle is recognized and detected, and the result of this detection collides. This is sent to the ECU for the determination process.

  By repeating the above processing, a robust voting method that is robust against noise without preparing prototype edge patterns for all vehicle types for all weather conditions such as image template matching pattern recognition processing while the vehicle 1 is traveling. By processing, the vehicle center position in the center in the vehicle width direction and the vehicle height direction of the preceding vehicle can be stably extracted as the vehicle feature amount of the preceding vehicle with a simple, low calculation amount, and an inexpensive and simple configuration. Based on this vehicle feature amount, the vehicle ahead can be recognized and detected stably and accurately.

  Further, since the image sensor is formed by the monocular camera 2 which is cheaper than the stereo camera, there is an advantage that it can be formed at a lower cost.

  And based on this vehicle detection result, the stable and accurate vehicle recognition, the judgment of the presence or absence of a collision, etc. can be performed, the reliability of advanced safety vehicles (ASV) etc. can be improved, and it can contribute to traffic safety.

  In the case of the above-described embodiment, the rectangular prediction region surrounding the vehicle ahead of the host vehicle is a region where the vertical and horizontal edges of the captured image of the monocular camera 2 are concentrated. Mounted with a distance sensor such as a wave radar, for example, the distance sensor scans in the left-right direction (vehicle width direction) and receives the reflected wave, so that the scanning position of the reflection point of the reflector of the vehicle ahead, It may be detected as a position in the vehicle width direction of the distance measurement target, and the prediction area may be set from these positions.

  The present invention is not limited to the above-described embodiment, and various modifications other than those described above can be made without departing from the spirit thereof. For example, the image sensor is limited to the monocular camera 2. It may be a stereo camera or the like.

  By the way, in order to reduce the number of equipment parts of the own vehicle 1, the monocular camera 2 and the like can also be applied to a case where it is also used as a sensor for other controls such as follow-up running control and brake control.

It is a block diagram of one embodiment. It is a flowchart for operation | movement description of FIG. FIG. 2 is an actual measurement diagram for explaining the processing of FIG. 1. It is explanatory drawing of the voting process by the side of the vertical edge of FIG. It is explanatory drawing of the voting process by the side of the horizontal edge of FIG. It is explanatory drawing of the synthetic | combination voting plane of FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Own vehicle 2 Monocular camera 5 Recognition process part Pi Image | photograph Pa Vertical edge image Pb Horizontal edge image Q Vehicle candidate area | region Ra Vertical edge side voting plane Rb Horizontal edge side voting plane Rab Composite voting top view

Claims (8)

In the captured image in front of the vehicle of the image sensor mounted on the vehicle, a rectangular prediction region surrounding the vehicle in front of the vehicle is set as a vehicle candidate region,
For the vertical edge image of the vehicle candidate area, vote +1 to the position where the vertical edge of the left half of the vehicle candidate area is moved to the right by the set length T *, and set each vertical edge of the right half of the vehicle candidate area The set length T * is defined as Tmin ≦ T * ≦ Tmax (Tmin and Tmax are the set minimum length and maximum length). ) Repeatedly while changing within the range of
For the horizontal edge image of the vehicle candidate area, vote +1 to the position where the horizontal edge of the lower half of the vehicle candidate area is moved upward by a set length H *, and set each horizontal edge of the upper half of the vehicle candidate area The set length H * is defined as Hmin ≦ H * ≦ Hmax, where Hmin ≦ H * ≦ Hmax (where Hmin and Hmax are the set minimum length and maximum). Long)
The vehicle center position in the center in the vehicle width direction and the vehicle height direction of the vehicle ahead of the host vehicle is extracted as a vehicle feature value from the coordinates of the peak value of the composite voting plane obtained by superimposing the voting results of the both edge side voting planes ,
A vehicle detection method for recognizing and detecting a vehicle ahead of the host vehicle from the vehicle feature amount.   2. The vehicle detection method according to claim 1, wherein the rectangular prediction region surrounding the vehicle ahead of the host vehicle is a region where vertical and horizontal edges of a captured image in front of the host vehicle are concentrated.   The distance prediction sensor such as a laser radar is mounted on the own vehicle, and a rectangular prediction region surrounding the vehicle ahead of the own vehicle is set from the position of the distance measurement target of the distance measurement sensor. Vehicle detection method.   The vehicle detection method according to claim 1, wherein the image sensor is a monocular camera. An image sensor that is mounted on the vehicle and photographs the front of the vehicle;
Candidate area setting means for setting a rectangular prediction area surrounding the vehicle ahead of the host vehicle as a vehicle candidate area in the captured image of the image sensor;
For the vertical edge image of the vehicle candidate area, vote +1 to the position where the vertical edge of the left half of the vehicle candidate area is moved to the right by the set length T *, and set each vertical edge of the right half of the vehicle candidate area The set length T * is defined as Tmin ≦ T * ≦ Tmax (Tmin and Tmax are the set minimum length and maximum length). ) Vertical edge side voting processing means that repeats while varying within the range of
For the horizontal edge image of the vehicle candidate area, vote +1 to the position where the horizontal edge of the lower half of the vehicle candidate area is moved upward by a set length H *, and set each horizontal edge of the upper half of the vehicle candidate area The set length H * is defined as Hmin ≦ H * ≦ Hmax (where Hmin and Hmax are set minimum lengths and maximums), by voting +1 at the position moved to the lower side of the length H * to form a horizontal edge side voting plane. Horizontal edge side voting processing means that repeats while being variable within a range of (long),
Voting result composition means for forming a composite voting plane obtained by superimposing the voting results of the both edge side voting planes;
Vehicle feature amount extraction means for extracting a vehicle center position in the center of the vehicle width direction and the vehicle height direction of the vehicle ahead of the host vehicle from the coordinates of the peak value of the composite voting plane as a vehicle feature amount;
A vehicle detection apparatus characterized in that a vehicle ahead of the host vehicle is recognized and detected from the vehicle feature amount.   6. The vehicle detection according to claim 5, wherein the candidate region setting means sets the rectangular prediction region surrounding the vehicle ahead of the host vehicle as a region where vertical and horizontal edges of the captured image in front of the host vehicle are concentrated. apparatus.   A distance measuring sensor such as a laser radar is mounted on the own vehicle, and the candidate area setting means sets a rectangular prediction area surrounding the vehicle ahead of the own vehicle from the position of the distance measuring target of the distance measuring sensor. The vehicle detection device according to claim 5, wherein   The vehicle detection device according to claim 5, wherein the image sensor is a monocular camera.