JPH07334800A - Vehicle recognition device - Google Patents

Abstract

PURPOSE:To provide a vehicle recognition device which measures the distance to a precedent vehicle by recognizing the precedent vehicle in the road image inputted by using an on-vehicle camera. CONSTITUTION:The analog video signal inputted by an image input means 1 is digitized by an A/D converting means 2 and stored in an image storage means 3. An edge extracting means 4 extracts edges from the digital image and then a lane area extracting means 5 is used to extract the lane on the road. Further, an edge projection means 6 and a symmetrical area extracting means 7 segment a rough presence candidate area of the precedent vehicle and an initial model setting means 8 sets an initial model. An outline extracting means 9 extracts the outline of the precedent vehicle by using the initial model which is thus set. A vehicle-to-vehicle distance calculating means 10 calculates the distance to the vehicle from the outline information on the precedent vehicle and a processing result output means 11 outputs the processing result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vehicle recognition device for recognizing a vehicle in front using image processing technology and measuring an inter-vehicle distance to the recognized vehicle.

[0002]

2. Description of the Related Art Conventionally, as a device for recognizing a front vehicle of this type, as described in, for example, Japanese Patent Application Laid-Open No. 1-281600, presence of a front vehicle by tracking an edge extracted from an image is known. There is one that extracts the area to be edited.

[0003]

However, in such a conventional vehicle recognition apparatus, it is said that the region cannot be accurately extracted when the vehicle edge extracted by the edge extraction processing has a discontinuity. I have a problem. In addition, in stereo distance measurement, there is generally a basic problem of how to find correspondence between images.

It is an object of the present invention to provide a vehicle recognition device capable of recognizing a front vehicle from a road image input using a video camera and measuring an inter-vehicle distance to the front vehicle.

[0005]

A vehicle recognition device of the present invention is a device for recognizing a preceding vehicle from a road image taken by a video camera of the own vehicle and measuring an inter-vehicle distance between the recognized vehicle and the own vehicle. There are a stereo image input unit mounted on a vehicle for photographing a road scene ahead, an A / D conversion unit for digitally converting an analog video signal input from the stereo image input unit, and the A / D conversion unit. Image storage means for storing a digitized road image; edge extraction means for subjecting the road image stored in the image storage means to differential processing to extract edge components;
Lane area extracting means for extracting a lane area from the road image stored in the image storage means, and edge components scattered in the lane area extracted by the lane area extracting means are projected on the horizontal axis or the vertical axis of the image. Edge projection means for extracting a candidate area for the presence of a forward vehicle from the road image stored in the image storage means, and extracting a left-right symmetrical area within the candidate area for the presence of the forward vehicle cut out by the edge projection means. By means of a symmetric area extracting means for further limiting the existence candidate area of the forward vehicle from the road image stored in the image storage means, and an initial contour model for the existence candidate area of the forward vehicle cut out by the symmetric area extracting means. Initial model setting means to be set, and stored in the image storage means based on the initial value set by the initial model setting means Contour extracting means for extracting the contour of the front vehicle from the road image being displayed, inter-vehicle distance calculating means for measuring the inter-vehicle distance to the front vehicle based on the contour shape of the front vehicle extracted by the contour extracting means, and the contour And a processing result output unit for outputting the vehicle contour extraction result extracted by the extraction unit and the inter-vehicle distance measured by the inter-vehicle distance calculation unit.

[0006]

According to this structure, the analog video signal input by the image input means is converted into a digital image by the A / D conversion means and stored in the image storage means. Then, after the edges are extracted from the digital image by the edge extracting means, the lanes on the road are extracted by using the lane area extracting means. Further, a rough existence candidate area of the vehicle ahead is cut out by the edge projection means and the symmetric area extraction means, and the initial model is set by the initial model setting means. The contour extracting means extracts the contour of the forward vehicle using the initial model set in this way.
The inter-vehicle distance calculating means calculates the inter-vehicle distance based on the contour information of the preceding vehicle, and the processing result output means outputs the processing result.

[0007]

Embodiments of the present invention will be described below with reference to the drawings. The vehicle recognition device of the present invention is configured as shown in FIG.

The image input means 1 for photographing the front of the vehicle is
As shown in FIG. 2, it is composed of a stereo camera composed of a pair of video cameras 1a and 1b. 2 is A / D
The conversion means digitizes the analog video signal input from the image input means 1. Reference numeral 3 is an image storage means, which is an A / D
The road image digitized by the conversion means 2 is stored. Four
Is an edge extraction means, which differentiates the road image stored in the image storage means 3 and stores it in the image storage means 3 again. Reference numeral 5 denotes a lane area extracting unit that extracts the own lane area in which the own vehicle is traveling and the adjacent lane area regions on the left and right sides thereof from the differential image accumulated in the image storage unit 3.
Reference numeral 6 denotes an edge projecting means, which roughly cuts out a vehicle presence candidate area by projecting an edge component extracted from the lane area on the vertical axis and the horizontal axis of the image. Reference numeral 7 denotes a symmetric area extracting means, which more accurately extracts the existence candidate area of the front vehicle by utilizing the left-right symmetry of the front vehicle. Reference numeral 8 denotes an initial model setting means, which applies an appropriate model to the presence candidate area of the forward vehicle cut out by the edge projection means 6 and the symmetrical area extraction means 7. Numeral 9 is a contour extracting means for finally extracting the contour of the forward vehicle by deforming the initial model set by the initial model setting means 8 based on the method of the active contour model. An inter-vehicle distance calculating means 10 calculates the inter-vehicle distance to the front vehicle by obtaining the parallax between the stereo images using the contour model of the front vehicle extracted by the contour extracting means 9. A processing result output unit 11 outputs the result of recognizing the front vehicle and the distance between the recognized front vehicle and the front vehicle.

The vehicle recognition apparatus shown in FIG. 1 is operated according to the processing procedure shown in FIG. First, in step 1000, the registers and counters are cleared, and various initial settings for repeating the subsequent processing are performed. Step
In 2000, the road image is captured. The road image is input as an analog video signal by using the pair of video cameras 1a and 1b, and the analog video signal is digitized by the A / D converter 2 and then stored in the image memory 3. 4A and 4B show a left input image and a right input image captured by the video cameras 1a and 1b.

In step 3000, a vehicle ahead is recognized from the road image captured in step 2000. Since the process of recognizing the front vehicle performs the same process for each of the pair of input images, only the process for the input image from the video camera 1b will be described below as an example with reference to FIG.

In the series of processes of the vehicle recognition process shown in FIG. 5, in step 3100, the step 200 shown in FIG.
Edges are extracted by applying differentiation processing to the road image captured at 0. Note that 3 × 3 Sobs are used for edge extraction.
The differential intensity obtained for each pixel is stored in the image storage means 3 using the el filter. FIG. 6 shows a result of _binarizing the differential intensity E (x, y) calculated for each pixel using the threshold value E _th . The threshold value E _th varies depending on the dynamic range of the video cameras 1a and 1b used for input, but is preferably set within the range of 80 to 120. Further, the points shown in black in FIG. 6 are pixels having a differential intensity exceeding the threshold value E _th , and hereinafter, this pixel will be referred to as an edge pixel.

In step 3200 shown in FIG. 5, a lane area is extracted from the road image. FIG. 7 shows lane area extracting means 5.
2 shows the flow of a series of processes in the above, and in this case, the processing region is limited to the lower half of the image in order to improve the accuracy of extracting the outline of the white line.

In step 3201 shown in FIG. 7, the contour of the left white line W1 painted on the road surface is extracted. Specifically, as shown in FIG. 8, pixels are scanned in the left direction from the center l of each scanning line. And the steps shown in FIG.
The differential intensity E (x, y) of each pixel obtained by 3100 is the threshold E
_The first position that exceeds _th is set as the contour point of the left white line in the scanning line. In Step 3202, the white line W2 on the right side
Extract the contour of. Similarly to the left side, the outline of the white line on the right side scans pixels in the right direction from the center l of each scanning line shown in FIG. 8, and the first when the differential intensity E (x, y) exceeds the threshold value E _th. The position is extracted as the contour point of the white line on the right side of the scanning line. FIG. 9 shows the result of extracting the left and right white line contour point sequences from the image. Pixels indicated by black dots are contour points W10 and W20 of left and right white lines.

In step 3203, the outline point sequence of the left white line extracted in step 3201 is linearly approximated. Step
Similarly, in 3204, the outline point sequence of the right white line extracted in step 3202 is linearly approximated.

A Hough transform method (US Pat. No. 3,069,654 (1962)) is used for the linear approximation processing of the point sequence performed in steps 3203 and 3204. In Step 3205, the straight line approximating the white line on the left side, the straight line approximating the white line on the right side, the bottom edge of the image, the left edge of the image,
The area formed by the right edge of the image is extracted as the lane area in which the vehicle is traveling. FIG. 10 shows the result of extracting the lane area in which the vehicle is traveling. Points V, L,
The area surrounded by B and A is the extracted lane area.

In step 3206, the left and right adjacent lanes adjacent to the traveling lane area of the own vehicle extracted in step 3205 are approximately obtained. In this case, as shown in FIG. 11, a triangle VCD in which the length of the bottom side LR of the triangle VLR representing the traveling lane area of the own vehicle is doubled to the left and right is extracted as the road area including the adjacent lane.

In step 3300 shown in FIG. 5, the existence candidate area of the preceding vehicle is obtained by examining the distribution of the edges scattered in the lane area extracted in step 3200. 12 and 13 are schematic diagrams showing the concept of this processing.

As shown in FIG. 12B, the processing area is limited to the lane area of the own vehicle extracted in step 3205 of FIG. Then, within this processing area, the threshold value E
Pixels with edge strength exceeding _th (called edge pixels)
Is counted for each scanning line to create a histogram as shown in FIG. At the same time, an average coordinate position of the scanning line direction of these edge pixels, the centroid position G _x in the scanning direction of the vehicle candidate region. The vertical axis of this histogram represents each scanning line, and the horizontal axis represents the number of edge pixels. Next, a threshold value _Bth is set for the number of edge pixels. As a result of conducting experiments using various images, it is preferable to set the threshold value B _{th to} about “40”. Then, of the scanning lines in which the number of edge pixels exceeds B _th , the scanning line located at the bottom is extracted as the lower end of the vehicle candidate area. In FIG. 12A, the scanning line indicated by B indicates the lower end of the vehicle candidate area. If a scanning line that satisfies the above conditions is not extracted, it is determined that there is no vehicle ahead.

When the lower end of the vehicle candidate area is extracted as described above, the left and right side edges of the vehicle candidate area are extracted. FIGS. 13A and 13B are schematic diagrams showing the concept of this processing.

As shown in FIG. 13A, the road area including the adjacent lanes extracted in step 3106 of FIG. 7, the lower end of the forward vehicle candidate area extracted in the previous process, and the left and right edges of the image are surrounded. The area to be processed is limited to the processing area. The reason for setting such a processing range is to constantly monitor the adjacent lane in preparation for the lane change of the vehicle ahead and the interruption of another vehicle. Then, in this processing area, the number of edge pixels is counted for each vertical pixel column perpendicular to the scanning line, and a histogram as shown in FIG. 13B is created. The horizontal axis of this histogram represents the horizontal coordinate of the image, and the vertical axis represents the number of edge pixels. In this case, as is clear from (a) and (b) of FIG. 13, in an area outside the area where the vehicle exists, as compared with the inside of the vehicle candidate area,
It can be seen that the frequency of the histogram sharply decreases and its variation decreases. Therefore, the left and right side edges of the vehicle candidate area are extracted according to the following procedure.

First, a small window having a section width S _W is provided on the horizontal axis of this histogram. Next, while calculating the average value E _mean and the variance value E _sigma of the edge frequencies within this small window, the window is set to the center of gravity G of the vehicle candidate region.
Shift from _x to the left and right. And
The first positions where E _mean is equal to or less than the threshold value M _th and E _sigma is equal to or less than S _th are extracted as the left end and the right end of the vehicle candidate region, respectively. The section width S _W of the small window is “20”, the threshold M _th for E _mean is “15”, E
_The threshold value S _th for _sigma is preferably set to about “17”. By the above processing, the vehicle candidate area N
FIG. 14 shows the result of extracting the.

In step 3400 of FIG. 5, a symmetric area extraction process is performed. In general, the forward vehicle shown in the image shows a substantially symmetrical shape with a line segment perpendicular to the scanning line as the axis of symmetry.
Therefore, a left-right symmetrical region is extracted within the range of the front vehicle candidate region N cut out in step 3300 to further limit the region where the front vehicle exists. The outline of these processes will be described with reference to FIGS.

First, as shown in FIG. 15, the processing range for extracting the symmetrical region is limited to the forward vehicle candidate region cut out in step 3300 of FIG. Next, a method for obtaining a symmetry axis perpendicular to the scanning line in this processing area will be described with reference to FIG. For example, it is assumed that a total of six edge points A, B, C, D, E, and F shown in FIG. 16 exist on a scanning line having the differential binary image in FIG. At this time, a histogram in which the midpoint positions between arbitrary two edge points among these six edge points are integrated is created, and the position where the frequency shows a peak is set as the axis of symmetry in the scanning line. Such processing,
This is performed for all the scanning lines in the area cut out by the edge projection means 6 to obtain the axis of symmetry of the entire area. Then, a set of edge points that are bilaterally symmetric with respect to the axis of symmetry thus obtained is extracted, and a bilaterally symmetric region is cut out. Figure 1
7 and FIG. 18 show a series of flow of this process.

First, the axis of symmetry is extracted by the processing from step 3401 to step 3412 shown in FIG. 17 and step 3413 shown in FIG. Step 3401 and Step 3402
Then, the initial value of the processing range is set. The processing range in FIGS. 17 and 18 is the vehicle candidate area cut out in step 3300 of FIG. 5, in which the upper limit y coordinate is Ty, the lower limit is By, the left limit x coordinate is Lx, and the right limit is. It is represented by Rx. Then, as shown in step 3403, the differential intensity E (x, y) at the coordinates (x, y) in the image is the threshold value E _th.
If it exceeds the threshold, as shown in steps 3404 to 3408, the point exists on the same scanning line and the threshold value E _th
The midpoints of all the pixels having a differential intensity exceeding .alpha. Are obtained, and in step 3406, the midpoints are added to the histogram corresponding to the midpoint positions. Note that this processing is performed in steps 3409, 3410, 3411, and 3412.
It is repeated for all edge points in the processing area, as shown in FIG. Then, in step 3413, the peak of the histogram thus obtained is obtained, and the x coordinate at that time is stored as the symmetry axis x _sym of the vehicle region. Furthermore, FIG.
In steps 3414 to 3423 shown in
The symmetric region about the symmetry axis x _sym thus obtained is extracted. This process is performed after initializing the processing area in Step 3414 and Step 3415, and then executing Step 341.
Pixel (x, y) whose differential intensity exceeds the threshold value E _th in 6
If it is confirmed, the distance D between the pixel and the axis of symmetry x _sym is determined in step 3417. And step 3
At 418, it is determined whether or not there is a pixel whose differential intensity exceeds the threshold value E _th at the position of the distance D with respect to the symmetry axis x _sym . If so, step through the resulting symmetry point pair
Register with 3419. In addition, this process is step 3420.
From step 3423, all edge points in the processing area are processed. The result of extracting the symmetrical region from the image of FIG. 15 in this way is shown in FIG. Note that FIG. 6B is a histogram created when the axis of symmetry is obtained.

Further, in step 3424, a rectangle circumscribing the symmetric region thus extracted is obtained. FIG. 20 shows a schematic diagram showing the concept of this processing. 20 (a) is shown in FIG.
This is the result of extracting a symmetric region as in (a) of 9. Then, as shown in (b) of FIG. 20, a histogram in which the edge points appearing in this image are projected on the horizontal axis of the image is created. The rightmost side is extracted as the right end of the symmetrical area. Furthermore, a histogram in which the edge points of (a) of FIG. 20 are projected on the vertical axis of the image is created as shown in (c) of FIG. 20. Of the histograms whose frequency exceeds the threshold, the uppermost scanning line is As the upper end of the symmetrical area, the lowermost scanning line is extracted as the lower end of the symmetrical area.
A rectangle circumscribing the edge of the symmetric region thus obtained is obtained as shown in FIG. 21, and the width of the rectangle is W and the height is H.
And

Next, in step 3500 of FIG.
Set the initial model of the active contour model for the symmetric region extracted by the 3400. The initial model has a shape that approximates the vehicle shape as shown in FIG. 22, and is composed of n nodes arranged at equal intervals. The number of nodes n in this case is preferably about 44. Then, the width W _m and height H _m of the initial model are calculated in step 34.
The parameter "W" and "H" obtained in 24
"By multiplying the set as shown in the diagram of FIG. 22. The parameters P" P preferably set in a range of 1.05 to 1.07 ". In addition, the center of gravity C _m and step 3400 of this initial model By installing the initial model so that the center of gravity C of the symmetric region obtained in
The initial value of the active contour model can be set at an appropriate position with respect to the vehicle area. FIG. 23 shows how the initial model Q is set in this way.

In step 3600 of FIG. 5, the active contour model is
The contour of the vehicle ahead is extracted using the method of Le. Dynamic contour
The model has an energy function based on the image features and model shape.
Number E _snakesAnd minimize this energy function
This is a method of extracting the contour of the object in the process. energy
Function E_snakesOf the model such as smoothness and distance between nodes
Internal energy E which is the force related to the shape_int , The model is
Image energy E, which is the force attracted to image features
_image , And the force to suppress the shape change of the model from the outside
External energy E_con And consists of
It is expressed as (1). v_i (I = 1,2,3, ..., n) is
It is a node of the contour model.

E _snakes (v _i ) = E _int (v _i ) + E _image (v _i ) + E _con (v _i ) ... (1) Further, the internal energy E _int can be calculated by the equation (2). it can. α and β are weighting parameters for each term.

E _int (v _i ) = α | v _i −v _i−1 | 2 + β | v _i−1 −2v _i + v _{i + 1} | 2 (2) Further, the image energy E _image is an image. The potential field from the inner edge is calculated as the density gradient of the image as shown in Expression (3). γ is a weighting parameter for image energy.

E _image (v _i ) = − γ | ▽ I (x, y) | (3) Further, as the external energy, the following formula (4) is taken into consideration in consideration of the left-right symmetry of the vehicle ahead. As shown in (3), a constraint force for shape change is applied so that the contour model contracts symmetrically. Note that g is the barycentric coordinates of the contour model, and v _i * is the symmetry point of v _i with respect to the axis of symmetry passing through the barycenter C. δ is a weight parameter for external energy.

E _con (v _i ) = δ || v _i −g | − | v _i * −g || (4) Then, the energy function E _snakes defined in this way
Is evaluated in the region near each node of the contour model, and the model is contracted by moving the node to the position where the energy is the smallest.

When the number of nodes to be moved becomes less than a certain threshold value, it is judged that the contour model has converged, and the contraction of the model is finished. FIG. 25 shows a flow of this processing.

First, step 3601 and step 360
Initialize the parameters in 2. Next, in step 3603, E _snakes is calculated according to equations (1) to (4).
In step 3604, E calculated in step 3603
Compare the _snakes with the energy of adjacent pixels. E
_{If snakes} are determined to be smaller, step 3
In step 605, E _snakes is held as the minimum value of energy and the parameters are added in step 3606 to calculate the energy of the next adjacent pixel, and then the process returns to step 3603. On the contrary, when it is determined that E _snakes is larger, the step is performed without updating the minimum value of energy.
Return to 3603. It should be noted that this process is repeated for all the preset neighboring regions based on the judgment in step 3607. When it is determined in step 3608 that the final minimum value of energy can be obtained at the current position of the node, the process directly proceeds to step 3610. On the contrary, if it is determined that the minimum value of energy is obtained in the adjacent pixel other than the current position of the node, in step 3609 the node is moved to the position of the adjacent pixel and step 3610
Go to. FIG. 24 shows the processing concept of this part.

Then, in step 3610, after adding the parameters, the processes of steps 3602 to 3611 are repeated for all nodes based on the judgment of step 3611. If it is determined in step 3611 that processing has been completed for all nodes, the number of moved nodes is evaluated in step 3612, and if this number is less than or equal to the threshold value, it is determined that the contour model has converged and processing is performed. To finish. Conversely, if the number of moving nodes is greater than or equal to the threshold,
Return to 3601 and repeat the process for all nodes. With the above processing, the contour of the vehicle ahead can be extracted from the road image. FIG. 26 shows a contour model R obtained as a result of convergence by extracting the contour of the forward vehicle.

Next, in step 4000 of FIG.
Measure the distance to the preceding vehicle recognized by the 3000. The measurement of the inter-vehicle distance is calculated according to the principle of triangulation by extracting the parallax between the front vehicles shown in the stereo images captured from the video cameras 1a and 1b.

27A and 27B, a contour model R1, which approximates the contour of the front vehicle from the input images from the left and right video cameras 1a and 1b of FIG. 4 by using the contour extracting means 9, respectively. R2 is shown.

FIGS. 28 (a) and 28 (b) show only the contour model among them. 29A and 29B are overlapped so that the origins of the stereo images coincide with each other, and the shift between the contour models R1 and R2 on each scanning line corresponds to the parallax on each scanning line. Then, a histogram is created in which the horizontal axis represents the parallax in each scanning line thus obtained and the vertical axis represents the number of scanning lines indicating the parallax, and the one with the largest parallax among those whose frequency exceeds the threshold is the front. Extract as corresponding to the vehicle.

Further, using the parallax thus obtained, the inter-vehicle distance is calculated according to the equation (5). Distance D = (camera optical axis distance DB x focal length f) / (parallax (pixel) d x pixel size PS) (5) The optical axis distance DB of the video cameras 1a and 1b is 1
m and the focal length f are preferably about 7.5 mm. Although the pixel size PS varies depending on the image sensor used, it is preferable to use a pixel having a resolution as high as possible.

In step 5000 of FIG. 3, the inter-vehicle distance to the preceding vehicle calculated by the above processing is output. Output the result on the display installed in the vehicle. An example of output to the display is shown in FIG. In addition, an auto-cruise device that keeps the vehicle-to-vehicle distance constant and traveling, or an inter-vehicle distance warning device that issues an alarm to the driver when the vehicle-to-vehicle distance becomes a certain value or less, calculated by the present invention. In order to realize various application functions using the following distance information,
An output terminal such as RS232C is provided.

In the above embodiment, the Sobel filter is used in the edge extraction processing from the image.
Any filter that can extract edges from an image such as Laplacian may be used. In the lane area extraction processing, when the straight line contour point sequence is approximated by Ho
Although the Ugh transformation is used, any linear approximation method such as the least square method may be used. In the embodiment of the present invention, the shape of a normal passenger car is used as the initial model of the active contour model, but the shape of another vehicle model such as a large truck may be used.

[0041]

As described above, according to the present invention, a vehicle ahead is recognized from a road image input using a video camera,
It provides a device that measures the distance between the recognized front vehicle and the front vehicle.The contour of the front vehicle is set by setting the initial model of the active contour model based on the edge distribution in the image and the left-right symmetry of the front vehicle. In addition to being able to extract, by calculating the parallax between the left and right stereo images using the contour model of the extracted front vehicle, it is possible to calculate the inter-vehicle distance with the front vehicle with high accuracy and at high speed. is there.

[Brief description of drawings]

FIG. 1 is a block diagram showing a basic configuration of an embodiment of the present invention.

FIG. 2 is a diagram showing an example of installation of a video camera in a vehicle.

FIG. 3 is a flowchart showing the operation of the embodiment of the present invention.

FIG. 4 is a diagram showing an example of input images from the cameras 1a and 1b.

FIG. 5 is a flowchart showing the operation of vehicle recognition processing.

FIG. 6 is a diagram showing a result of extracting edges from an input image.

FIG. 7 is a flowchart showing the operation of lane area extraction processing.

FIG. 8 is a schematic diagram showing a process of extracting the outline of a white line.

FIG. 9 is a diagram showing a result of extracting outlines of white lines.

FIG. 10 is a diagram showing a result of extracting a lane area.

FIG. 11 is a diagram showing a method of extracting adjacent lanes.

FIG. 12 is a schematic diagram showing a concept of a process of extracting a lower end of a vehicle candidate region in the edge search process.

FIG. 13 is a schematic diagram showing the concept of processing for extracting the left and right ends of a vehicle candidate area in the edge search processing.

FIG. 14 is a diagram showing a result of extracting vehicle candidate regions by edge search processing.

FIG. 15 is a diagram showing a processing range in symmetrical area extraction processing.

FIG. 16 is a schematic diagram showing a processing concept of symmetrical area extraction processing.

FIG. 17 is a flowchart showing the operation of symmetrical area extraction processing.

FIG. 18 is a flowchart showing the operation of symmetrical area extraction processing.

FIG. 19 is a diagram showing a processing result of symmetrical area extraction processing.

FIG. 20 is a schematic diagram showing a method of obtaining a region edge of a symmetrical region.

FIG. 21 is a diagram showing a result of obtaining a rectangle circumscribing a symmetric region.

FIG. 22 is a diagram showing an initial model of an active contour model.

FIG. 23 is a diagram showing a setting state of an active contour model.

FIG. 24 is a diagram showing a method of minimizing energy of an active contour model.

FIG. 25 is a flowchart showing the operation of contour extraction processing by the method of the active contour model.

FIG. 26 is a diagram showing a vehicle contour extraction result by the contour extraction processing.

27 is a view showing a result of extracting the contour of the vehicle from the input image shown in FIG.

28 is a contour model diagram of the contour extraction result shown in FIG.

[Fig. 29] Fig. 29 is a diagram illustrating a parallax calculation method for both contour models of a stereo image.

FIG. 30 is a diagram showing a display example of processing results.

[Explanation of symbols]

 1 image input means 1a video camera 1b video camera 2 A / D conversion means 3 image storage means 4 edge extraction means 5 lane area extraction means 6 edge projection means 7 symmetric area extraction means 8 initial model setting means 9 contour extraction means 10 inter-vehicle distance Calculation means 11 Processing result output means

Continuation of front page (51) Int.Cl. ⁶ Identification code Office reference number FI Technical display area H04N 7/18 C

Claims (8)

[Claims] 1. A device for recognizing a front vehicle from a road image captured by a video camera of the own vehicle and measuring an inter-vehicle distance between the recognized vehicle and the own vehicle. A stereo image inputting means for photographing the image, an A / D converting means for digitally converting an analog video signal input from the stereo image inputting means, and an image storing means for storing a road image digitized by the A / D converting means. An edge extracting means for extracting an edge component by differentiating the road image stored in the image storage means, and a lane area extracting means for extracting a lane area from the road image stored in the image storage means. The edge components scattered in the lane area extracted by the lane area extracting means are stored in the image storage means by projecting them on the horizontal axis or the vertical axis of the image. Edge projection means for extracting the presence candidate area of the forward vehicle from the existing road image, and stored in the image storage means by extracting a left-right symmetrical area within the presence candidate area of the forward vehicle cut out by the edge projection means. Symmetric region extracting means for further limiting the presence candidate region of the forward vehicle from the road image, and initial model setting means for setting an initial contour model for the presence candidate region of the forward vehicle cut out by the symmetric region extracting means, A contour extracting means for extracting the contour of the front vehicle from the road image stored in the image storage means based on the initial value set by the initial model setting means; and a contour shape of the front vehicle extracted by the contour extracting means. An inter-vehicle distance calculating means for measuring an inter-vehicle distance based on a vehicle ahead, and a vehicle wheel extracted by the contour extracting means. Extraction result and the vehicle recognition apparatus provided with a processing result output means for outputting the inter-vehicle distance measured by the inter-vehicle distance calculating means. 2. The lane area extracting means extracts the lane area in which the host vehicle is traveling from the road image stored in the image storage means, and the lane area is equal to the left and right lane widths shown at the bottom of the image. The vehicle recognition device according to claim 1, wherein the vehicle recognizing device is configured to approximately extract the left and right adjacent lane areas by extending the area by a factor of two. 3. After extracting the lower end of the forward vehicle candidate area by projecting the edge component in the lane area in which the host vehicle is running, which is extracted by the lane area extracting means, on the vertical axis of the image, the edge projecting means The left and right ends of the forward vehicle candidate region are extracted by projecting the edge components in the lane region in which the vehicle is traveling and the adjacent lane region extracted by the lane region extracting means onto the horizontal axis of the image. 1, claim 2
The vehicle recognition device described. 4. The vehicle recognition device according to claim 3, wherein the edge projection means is configured to extract the area edge of the vehicle candidate area based on the variance value of the number of edges extracted from the lane area. 5. The symmetric region extracting means is configured to extract the symmetry axis of the entire region by accumulating the midpoint positions obtained between arbitrary edge points existing on the same scanning line. Vehicle recognition device. 6. The initial model setting means is configured to obtain the shape and size of a model adapted to the symmetric area based on the area width and height of the symmetric area extracted by the symmetric area extracting means. Vehicle recognition device. 7. The initial model setting means obtains the shape and size of the model adapted to the symmetric area based on the area width and height of the symmetric area extracted by the symmetric area extracting means, and extracts by the symmetric area extracting means. The vehicle recognition device according to claim 1, wherein the model adapted to the symmetric region is automatically set to an appropriate position by matching the barycentric position of the symmetric region with the barycentric position of the model. 8. An inter-vehicle distance calculating means uses a contour model of a front vehicle extracted from one of input images obtained from a right camera and a left camera as a reference pattern, and a contour of a front vehicle extracted from the other input image. The vehicle recognition device according to claim 1, wherein the inter-vehicle distance to the preceding vehicle is calculated by obtaining a parallax between the model and the model.