JPH0916782A - Object recognizing device - Google Patents

Abstract

PURPOSE: To recognize an object with high accuracy by expanding an edge point extracted from image pickup data for the unit of a small area divided on a grid. CONSTITUTION: The image pickup area of a CCD camera 12 is divided in the shape of grid and expansion processing of the longitudinal and lateral edge points is performed for the unit of each small area. When the longitudinal edge point exists inside, each small area divided like this is grasped as an area expressing the longitudinal edge of the object existent in the photographic data and when the lateral edge point exists inside, the small area is grasped as an area expressing the lateral edge of the object existent in the photographic data. Then, the linear longitudinal edge area corresponding to the longitudinal edge of a preceding vehicle and the linear lateral edge area corresponding to the lateral edge of the preceding vehicle are respectively continuously and sharply formed in a longitudinal edge area picture. Therefore, the angle point of the object is specified based on the area where these longitudinal edge area group and lateral edge area group cross, the presence/absence of the preceding vehicle is judged and the size or the like of the preceding vehicle is detected.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an object recognition device,
In particular, the present invention relates to an object recognition device suitable as a device used for recognizing a preceding vehicle in an automobile.

[0002]

2. Description of the Related Art Conventionally, a device for recognizing an object based on image data of an image sensor has been known. In such an object recognition device, a method is generally used in which vertical edges and horizontal edges are detected from imaged data and an object is recognized based on the obtained edge shape. The vertical edge and the horizontal edge in the imaged data are detected by comparing the luminances of the obtained imaged data on a pixel-by-pixel basis and determining whether or not there is a luminance difference exceeding a predetermined value between adjacent pixels.

Specifically, the brightness of each pixel of the image sensor is compared between adjacent pixels in the horizontal direction, and if there is a brightness difference exceeding a predetermined value, those pixels are regarded as vertical edge points. Be recognized. When the vertical edge points are continuously detected in the vertical direction, the set of them is recognized as a vertical edge. Similarly, the brightness in each pixel of the image sensor is compared between pixels adjacent in the vertical direction, and when there is a brightness difference exceeding a predetermined value, those pixels are recognized as horizontal edge points, and further, When horizontal edge points are continuously detected in the horizontal direction, a set of them is recognized as a horizontal edge.

When an object is recognized using the above method, the contour of the object may not be obtained as a clear linear edge, but may be detected as a discontinuous edge. When the contour of the object is detected as a discontinuous edge in this way, it is difficult to distinguish between the edge representing the contour of the object and the edge detected corresponding to a portion other than the contour of the object, and A situation occurs in which it is not possible to recognize various objects.

As an object recognition apparatus focusing on such a problem, as disclosed in Japanese Patent Laid-Open No. 62-126499, when an edge point is detected, several pixels adjacent to the pixel detected as the edge point are adjacent. There is known a device that expands the obtained edge by recognizing, for example, adjacent 4 pixels or 8 pixels as an edge region. According to such a device, even when the contour of the object is obtained as discontinuous edge points, since these edge points are respectively expanded, a continuous edge region can be obtained. Therefore, according to the above object recognition device, higher object recognition accuracy can be obtained as compared with the case where the object is directly recognized from the edge point.

[0006]

However, when the edge points are detected based on the image pickup data, there are cases where a plurality of vertical edges are detected close to each other in the horizontal direction. Also,
Similarly, a plurality of horizontal edges may be detected close to each other in the vertical direction. Under such a circumstance, if the edge expansion process is performed on each edge point, the vertical edge region and the horizontal edge region may become unreasonably thick regions.

When the vertical edge region and the horizontal edge region are detected in the imaged data and the object is recognized based on these edge regions, in order to accurately recognize the contour of the object, it corresponds to the contour. It is desirable that the edge region to be formed is thin. In this respect, the above-mentioned conventional object recognition device in which a thick edge region may be set unnecessarily has a problem that it is not possible to always maintain good object recognition accuracy.

The present invention has been made in view of the above points, and provides an object recognizing device that solves the above problems by expanding an edge detected in imaged data within a predetermined expansion width. The purpose is to do.

[0009]

In the object recognition device for recognizing an object using an image sensor, the image pickup area of the image sensor is formed into a grid pattern in units of small areas including a predetermined number of pixels. Area dividing means for dividing, edge point detecting means for detecting vertical edge points and horizontal edge points in imaged data, a small area including the vertical edge points, a vertical edge area, and a small area including the horizontal edge points An edge point for specifying the corner point of the object based on an area where an edge area setting unit having a horizontal edge area, a vertical edge area group continuous in the vertical direction, and a horizontal edge area group continuous in the horizontal direction intersects each other. It is achieved by an object recognition device comprising: specifying means.

[0010]

In the present invention, the image sensor captures an image including an object to be recognized. The edge point detecting means detects a vertical edge point and a horizontal edge point from the imaged data captured by the image sensor. When an object to be recognized is imaged in the imaged data, a plurality of vertical edge points and a plurality of horizontal edge points are detected substantially along the contour of the object.

The area dividing means divides the image pickup area of the image sensor into a grid pattern. When a vertical edge point exists in each of the small areas divided in a grid pattern, the edge area setting means sets the small area as a vertical edge area. Further, when a horizontal edge point exists in each of the small areas divided in a grid pattern, the edge area setting means sets the small area as a horizontal edge area. The vertical edge area and the horizontal edge area always have the grid width of the small area even if a plurality of edge points exist in each small area. Therefore,
The width of the edge region is not unduly widened even in a region where a plurality of vertical edge points are detected close to each other and a region where a plurality of horizontal edge points are detected close to each other.

The corner point specifying means recognizes vertical edge areas continuous in the vertical direction as a vertical edge area group, and recognizes horizontal edge areas continuous in the horizontal direction as a horizontal edge area group,
Further, the corner points of the object are specified based on the region where the vertical edge region group and the horizontal edge region group intersect. When the object to be recognized is imaged in the imaged data, the vertical edge regions are continuously detected in the vertical direction and the horizontal edge regions are continuously detected in the horizontal direction corresponding to the contour of the object. Therefore, the vertical edge region group and the horizontal edge region group recognized by the corner point specifying means correspond to the contour of the object to be imaged, and the intersections thereof correspond to the corner portions of the object.

[0013]

1 is a block diagram of an object recognition apparatus according to an embodiment of the present invention. The object recognition device of this embodiment is a device configured to recognize a preceding vehicle in an automobile.

The object recognition apparatus of this embodiment is equipped with a radar 10 and a CCD camera 12 which have a monitoring area in front of the vehicle. The radar 10 is a device for measuring the distance to an object existing in front of the vehicle, and its output signal is supplied to a bus line 18 leading to a digital signal processor (DSP) 16 via an A / D converter 14. ing.

The CCD camera 12 is a camera for picking up an image of the front of the vehicle, and takes in an image including the preceding vehicle as image pickup data. The output signal of the CCD camera 12 is supplied to the input video RAM 22 via the A / D converter 20. The input video RAM 22 is connected to the bus line 18, stores the image pickup data captured by the CCD camera 12, and provides the image pickup data on the bus line in accordance with a command from the DSP 16.

The bus line 18 has a main R in addition to the A / D converter 14 and the input video RAM 22 described above.
The AM 24, the ROM 26, the output video RAM 28, and the data RAM 30 are connected. Main RAM 24
Is a memory for storing the image data after the DSP 16 processes the image captured from the input video RAM 22. In this embodiment, since it is necessary to store a plurality of image data in the process of processing, the main RAM
24 is provided with a memory capacity for storing a predetermined number of image data. The ROM 26 is a memory for storing a program to be executed by the DSP 16.
The object recognition apparatus of this embodiment is characterized by the contents of the processing executed by the DSP 16 according to the program stored in the ROM 26. Regarding the processing contents,
I will explain in detail later.

The output video RAM 28 is a memory for storing an image finally obtained in order to determine whether or not there is a preceding vehicle. The output video RAM 28 has a D / A
A video display 34 is connected via the converter 32. The video display 34 is a display provided inside the vehicle compartment to display various kinds of information to an occupant of the vehicle. The DSP 16 performs various processes on the image data captured by the CCD camera 12 to form a final image for determining the presence / absence of a preceding vehicle, stores the image data in the output video RAM 28, and further stores the image data. Is displayed to the occupants of the vehicle via the video display 24.

The data RAM 30 stores various data that can be detected from an image finally obtained to determine the presence or absence of a preceding vehicle, such as data regarding the presence or absence of a preceding vehicle and data regarding the size of the preceding vehicle. Is the memory of. The data RAM 16 is connected to a vehicle control device 36 including an automatic braking device, an alarm device and the like. The vehicle control device 36 detects the traveling environment of the own vehicle using the data stored in the data RAM 16 and executes predetermined vehicle control according to the state.

FIG. 2 shows an example of image pickup data taken from the CCD camera 12. In the imaging data shown in FIG. 2, a preceding vehicle 40 traveling in front of the vehicle, a road surface 42 drawing a gentle left curve, a road white line 44 drawn on the road surface, a guardrail 46 extending to the right of the road surface 42, Preceding vehicle 40
A large vehicle 48 and the like running in parallel are included. Various objects existing in the image data include the color, surface state,
An image is captured with different brightness depending on the angle between the object and the CCD camera 12. Therefore, when a point having a large difference in brightness between adjacent pixels in the image pickup data is detected, the contours of various objects existing in the image pickup data can be extracted.

FIG. 3 is image data obtained by scanning the imaged data shown in FIG. 2 in the horizontal direction of the screen and extracting pixels having a large luminance difference exceeding a predetermined value between horizontally adjacent pixels as edge points. Indicates. As shown in FIG. 3, when the edge point is detected by such a method, the edge extending in the vertical direction of the object existing in the imaging data is detected as the edge point. Hereinafter, as shown in FIG. 3, an edge point representing a vertical edge of an object is referred to as a vertical edge point, and a screen displaying the vertical edge point is referred to as a vertical edge screen.

In addition, in FIG. 4, the image pickup data shown in FIG. 2 is scanned in the vertical direction of the screen, and pixels having a large luminance difference exceeding a predetermined value between adjacent pixels in the vertical direction are extracted as edge points. Image data is shown. As shown in FIG. 4, when the edge point is detected by such a method, the edge extending in the lateral direction of the object existing in the imaging data is detected as the edge point. Hereinafter, as shown in FIG. 4, an edge point representing an edge of the object in the horizontal direction is referred to as a horizontal edge point, and a screen displaying the horizontal edge point is referred to as a horizontal edge screen.

The outline of the vehicle as viewed from the rear includes a straight line extending in the vertical direction and a straight line extending in the horizontal direction. Therefore, as shown in FIG.
When the vehicle is traveling, in the vertical edge screen shown in FIG. 3, a plurality of vertical edge points are used to extend straight lines extending in the vertical direction at positions corresponding to both ends of the preceding vehicle 40 (hereinafter,
As long as 40 V is detected (referred to as a vertical edge straight line),
In the horizontal edge screen shown in, by multiple horizontal edge points,
A plurality of straight lines (hereinafter, referred to as lateral edge straight lines) 40H extending in the vehicle width direction of the preceding vehicle 40 are detected. On the other hand, when the preceding vehicle 40 does not exist in front of the own vehicle, the vertical edge straight line 4
The 0V and the horizontal edge straight line 40H are not detected in front of the vehicle.

Therefore, the vertical edge screen and the horizontal edge screen are formed by using the image pickup data captured from the CCD camera 12, and a suitable number of vertical edge straight lines 40V and horizontal edge straight lines are formed at appropriate positions in these screens. By determining whether 40H is detected, the preceding vehicle 40
It is possible to determine whether or not exists.

However, the vertical edge straight line 40 of the preceding vehicle 40
The V and the horizontal edge straight line 40H cannot always be detected as an ideal state, that is, a straight line continuous in the vertical direction and a straight line continuous in the horizontal direction, and as a discontinuous straight line depending on the traveling environment of the vehicle. May be detected. In front of the vehicle, there are various objects other than the preceding vehicle 40,
Edge straight lines representing the contours of these objects are also detected in the vertical edge screen and the horizontal edge screen. Therefore, if the vertical edge straight line 40V and the horizontal edge straight line 40H are discontinuous, it is not always easy to determine whether they represent the contour of the preceding vehicle 40 or the contour of another object. . Therefore, the screen data of the vertical edge screen,
And the preceding vehicle 4 directly from the screen data of the side edge screen
If the presence / absence of 0 is determined, good detection accuracy may not be maintained in some cases.

On the other hand, an image obtained by forming vertical edge screens and horizontal edge screens as shown in FIGS. 3 and 4 and then expanding the vertical edge points and horizontal edge points detected in those screens. If data is formed, vertical edge straight line 40V,
It is possible to obtain new image data in which the discontinuity points of the horizontal edge straight line 40H are continuous. Therefore, if the presence or absence of the preceding vehicle 40 is determined based on the image data,
It is possible to always maintain good detection accuracy.

However, if all vertical edge points and horizontal edge points are uniformly expanded, for example, when a plurality of vertical edge points are detected in the horizontal direction, or the horizontal edge points are detected in the vertical direction. If a plurality of points are detected in the area, the width of the area representing the contour of the preceding vehicle 40 in the new image data is unnecessarily enlarged. In this case, it becomes difficult to accurately grasp the size of the preceding vehicle 40 based on the new image data, and there arises a problem that the detection accuracy of the data required by the vehicle control device 36 cannot be maintained satisfactorily.

On the other hand, in this embodiment, as shown in FIG.
As shown in FIG. 5, the image pickup area of the CCD camera 12 is divided into a lattice shape, and the vertical edge points and the horizontal edge points are expanded in units of each small area 12-n. Each of the small areas 12-n shown in FIG. 5 is set so that a predetermined number of pixels is secured inside. Each small area 1 divided in this way
2-n is a region representing a vertical edge of an object existing in the imaged data when a vertical edge point is present therein, and a horizontal edge point is present inside thereof.
It is grasped as an area representing a lateral edge of an object existing in the imaged data.

Therefore, as a result of detecting the vertical edge points, for example, as shown in FIG. 5, the pixel at the right end of the small area 12-k, the pixel at the left end of the small area 12-k + 1, and the small area 12-k. When two pixels at the left and right ends of +2 are detected as vertical edges, three small areas 12-k to 12-k + 2 that are vertically adjacent to each other are detected.
Will be recognized as a vertical edge region.

At this time, although the vertical edge points detected in the small area 12-k and the vertical edge points detected in the small area 12-k + 1 are laterally offset, No offset is generated in the vertical edge region resulting from these two vertical edge points. That is, according to the method of the present embodiment, when the vertical edge and the horizontal edge of the preceding vehicle 40 have a slight curvature, the linear vertical edge area group and the horizontal edge area group are formed on the image data. Will be obtained.
In addition, in the above example, the width of the vertical edge region is not expanded even though a plurality of vertical edge points offset in the horizontal direction are detected in the small region 12-k + 2. That is,
According to the method of the present embodiment, even when the edge points are detected adjacent to each other, the expansion width of the edge is restricted to the grid width of the small area 12-n, and the unreasonable expansion of the edge area width is prevented. It will be.

FIG. 6 shows a vertical edge area screen obtained by subjecting the vertical edge screen shown in FIG. 3 to the expansion processing described above. Further, FIG. 7 shows a horizontal edge area screen obtained by performing the expansion processing described above on the horizontal edge screen shown in FIG. As shown in FIGS. 6 and 7, in the vertical edge area screen, a linear vertical edge area 40Vw corresponding to the vertical edge of the preceding vehicle 40 is a linear horizontal edge area 40Hw corresponding to the lateral edge of the preceding vehicle 40. Are formed continuously and sharply. Therefore, if the presence or absence of the preceding vehicle 40 is determined based on the vertical edge area screen and the horizontal edge area screen and the size of the preceding vehicle is detected, the determination accuracy and the detection accuracy are always good. Can be maintained.

Hereinafter, with reference to FIGS. 8 to 18, an example of the processing executed by the object recognition apparatus of this embodiment for object recognition by the above method will be described. FIG.
4 is a flowchart of a main routine executed by the DSP 16 when recognizing an object using the image pickup data captured from the CCD camera 12. 9 to 11 are flowcharts of subroutines executed by the DSP 16, respectively. In the object recognition apparatus of this embodiment, the DSP 16 executes object recognition by executing the main routine shown in FIG. 8 and the subroutine shown in FIGS. 9 to 11.

When the main routine shown in FIG. 8 is started,
First, in step 100, from the image data captured by the CCD camera 12, that is, the input video RAM.
A process of extracting diagonal line segments from the imaged data stored in 22 is executed. In the captured image of the CCD camera 12, a white line on the road, a guardrail, or the like, which represents the traveling lane of the vehicle, is usually captured. In this step, in order to extract these contours, a process of extracting an oblique line segment that inclines in the upward right direction of the screen and an oblique line segment that inclines in the downward right direction of the screen from the imaged data is performed.

The process of step 100 is specifically shown in FIG.
It is performed by the subroutine shown in. When the routine shown in FIG. 9 is started, first, at step 200, C
An image in front of the vehicle captured by the CD camera 12 is read into the input video RAM 22. Next, in step 202, the area and the direction for scanning the imaging data are set. The CCD camera 1 is used to extract diagonal line segments.
The entire imaging area of 2 is performed as a scanning area.
Further, the scanning direction is set to the screen lower right direction when extracting the line segment in the screen right upward direction, and is set to the screen right upward direction when extracting the line segment in the screen lower right direction. .

After the scanning area and direction are set as described above, in step 204, edge points existing in the specified area are extracted and an edge image is created. In this embodiment, the CCD camera 12
The brightness of each pixel is classified into 256 levels, and when there is a brightness difference exceeding a predetermined value, for example, a brightness difference exceeding 200 levels between two pixels adjacent in the scanning direction, the pixel is captured as an edge point. I have decided. When the above process is completed, the edge image as shown in FIG. 12, for example, may be obtained by the process of extracting the line segment to the right, and the edge image shown in FIG.
Edge images as shown in are respectively created. The edge image created in this way is temporarily
It is stored in M24.

Next, at step 206, processing for dividing the edge image into small areas is performed. When the object existing in front of the own vehicle is the preceding vehicle 40, the small area is such that the vertical edges and the horizontal edges thereof are included in one small area row.
It is set according to the distance between the object existing in front and the own vehicle. Specifically, the width of the small area is set large when the distance to the front object is short, and the width of the small area is set small when the distance to the front object is long. The distance between the front object and the vehicle is measured based on the output signal of the radar 10.

When the process of dividing the edge image into small areas is completed, the maximum edge strength (luminance difference between adjacent pixels) existing inside each small area is calculated in step 208. And step 21 that follows
At 0, a small area image is created by filling each small area with the respective maximum edge strength.

After the above processing is completed, the routine proceeds to step 212, where the effective line segment is extracted from the small area image. In the small area image, in addition to the area corresponding to the contour of the object to be originally recognized, the contours of various objects are superimposed as noise. Therefore, in the present embodiment, only when a predetermined number or more of small areas having a large edge strength are continuously detected in the direction perpendicular to the scanning direction,
It is decided that these small area groups are the contours of the object to be recognized, and the line segment consisting of these small area groups is extracted as an effective line segment. Therefore, the series of small region groups extracted in step 212 clearly represent the outline of the object to be recognized.

The process of step 212 is specifically shown in FIG.
It is executed according to the subroutine shown in 0. When the routine shown in FIG. 10 is started, first in step 300, the scanning direction is set. In this step,
The same direction as the scanning direction set in step 202 is set as the scanning direction for extracting the effective line segment.

In step 302, it is judged whether or not the small area to be scanned has already been recognized as a small area exceeding the reference value (hereinafter referred to as a line segment forming small area). As a result, if it is determined that the segment is already recognized as the line segment configuration small area, steps 304 to 31.
2 is jumped, and then the process of step 314 is executed. On the other hand, when it is determined that the small area is not yet recognized as the line segment forming small area, the process of step 304 is executed next.

At step 304, it is judged if the edge strength in the small area exceeds the reference value. As a result, when it is determined that the edge strength in the small region does not exceed the reference value, it is determined that the small region is not the line segment-constituting small region, and the process proceeds to step 314. On the other hand, when it is determined that the edge strength in the small area exceeds the reference value, it is determined that the area is the line segment configuration small area, and the process proceeds to step 316.

In step 316, the target of the edge strength inspection is changed from the small area in which the existence of the edge strength exceeding the reference value is confirmed to the small area adjacent in the direction perpendicular to the scanning direction set in step 300. To be done. Next, at step 318, it is judged whether the edge strength in the small area newly investigated exceeds the reference value, that is, whether it is a line segment forming small area. As a result, when the edge strength in the small area exceeds the reference value, it is determined that the area is also the line segment forming small area, and the process returns to step 306. After that, the processes of repeating steps 306 and 308 are executed. On the other hand, when it is determined in step 308 that the edge strength in the newly investigated small area does not exceed the reference value, it is determined that the small area is not a line segment-constituting small area, and the following steps are performed. The process of 310 is executed. Therefore, when one line segment forming small area is detected in the process of scanning, the processes of the repeating steps 306 and 308 are repeated as long as the line segment forming small areas are continuous in the direction perpendicular to the scanning direction. When it is executed and the line segment constituent small areas become discontinuous, the process of step 310 is executed.

In step 310, it is judged whether or not the continuous length of the line segment forming small area detected by the above processing exceeds the reference value. The continuous length of the line segment structure small areas is relatively long when the line segment structure small areas are the contours of the target object to be recognized, while it is relatively short when they are not the contours of the target object to be recognized. Become. Further, the reference value used in this step 310 is such that, when the line segment forming small area is the contour of the target object to be recognized, its continuous length is longer than the reference value, and the line segment forming When the small area is not the contour of the target object to be recognized, the continuous length is set to be shorter than the reference value. Therefore, when the continuous length exceeds the reference value, it can be determined that the line segment forming small areas are the contours of the object to be recognized. In this case, the line segment constituent small areas are grasped as effective line segments, and thereafter, in step 312, the starting point and the length of the line segment are recorded, and then the process proceeds to step 314. On the other hand, if it is determined in step 310 that the continuous length does not satisfy the reference value, those line segment constituent small areas are not grasped as effective line segments, and the process of step 314 is directly executed thereafter.

At step 314, it is judged if scanning has been completed for the entire image pickup area of the CCD camera 12. As a result, if it is determined that scanning is not yet performed on all the areas, step 316
The process proceeds to step 302, and after changing one small area in which the edge strength is to be detected in the scanning direction, the process returns to step 302. Therefore, thereafter, the processes in and after step 302 are repeatedly executed until it is determined in step 314 that the scanning has been completed for all the regions. On the other hand, if it is determined in step 314 that scanning has been completed for all areas, the processing of this routine is ended at that point.

When the subroutine shown in FIG. 10 is completed,
In the subroutine shown in FIG. 9, the process of step 214 is executed. In step 214, the effective line segment extracted as described above is stored in the main RAM 24. When an oblique line segment having an inclination to the right of the screen and an oblique line segment having an inclination to the right of the screen, which are present in the image data, are detected by the subroutine shown in FIG. As shown, an image in which a line segment corresponding to the boundary line of the traveling lane of the own vehicle is extracted as an effective line segment is stored.

The subroutine shown in FIG. 9 is ended after the processing of step 214 is completed. After that, as shown in FIG.
The processing of the main routine shown in is restarted. In the main routine shown in FIG. 8, the process of step 102 is then executed. In step 102, based on the diagonal line segment extracted as described above, that is, based on the boundary line of the vehicle,
A process of identifying a region where a preceding vehicle is predicted to exist is performed.

The area where the preceding vehicle is to be detected in the vehicle is limited to the traveling lane of the own vehicle. For this reason, in this routine, the area between the two oblique direction line segments extracted as described above is limited as the area where the preceding vehicle exists. Therefore, when the diagonal line segment is extracted as shown in FIG.
The hatched area in FIG. 15 is specified as the area where the preceding vehicle exists.

Next, in step 104, a process of extracting a line segment in the vertical direction from the image data of the CCD camera 12 in the area specified as described above is performed. If a preceding vehicle exists in front of the host vehicle, the contour of the preceding vehicle is captured in the captured image data. In this step, a process of extracting an edge in the vertical direction of the contour is performed. The process of this step is
Specifically, it is executed by the above-mentioned subroutine shown in FIGS. 9 and 10. In this case, step 20 in FIG.
In 2, the space between the two oblique line segments extracted in the image data stored in the main RAM 24 is set as the scanning area, and in step 202 and step 300 in FIG. 1
The left and right directions of the second imaging area are set as the scanning direction.

According to this setting, when the image data as shown in FIG. 15 is obtained, for example, as shown by the arrow in FIG. Ongoing scanning is performed. At this time, if a large luminance difference exists between pixels adjacent in the horizontal direction in the imaged data, the point is captured as an edge point. Therefore, if the preceding vehicle exists in the area to be scanned, the vertical edge of the preceding vehicle is detected as the effective line segment, and in step 214 shown in FIG. 9, the vehicle width is changed as shown in FIG. The edge image data in which two effective line segments are extracted is separated from the main RAM 2 by a proper distance in the direction.
4 will be recorded.

When the extraction of the vertical line segment is completed, the main routine shown in FIG. 8 proceeds to step 106, and two vertical line segments that can be estimated to be vertical edges of the vehicle are extracted. A process of selecting a set of vertical line segments is executed. In this routine, the starting point and the ending point of the two vertical line segments are not largely offset in the vertical direction on the screen data, and the two vertical line segments have an appropriate distance in the vehicle width direction. A set of two vertical line segments that satisfy the above two points are selected as vertical edge candidates of the preceding vehicle.

When the above processing is completed, next step 1
At 08, a process of extracting a horizontal line segment existing in the imaging data is executed. If there is a preceding vehicle ahead of the host vehicle, the lateral edge of the preceding vehicle should be captured in the captured image data. If the set of two vertical line segments selected in step 106 is the vertical edge of the preceding vehicle,
The horizontal edge detected in the imaged data should be detected so as to connect these two vertical line segments. Therefore, in this routine, a process of extracting a horizontal line segment passing through each vertical edge set selected in step 106 is performed.

The processing of this step is executed by the subroutine shown in FIGS. 9 and 10 described above, similarly to the processing of steps 100 and 104. In this case, FIG.
In the middle step 202, an appropriate area including all the selected vertical line segments is set as the scanning area, and in the same step 202 and step 300 in FIG. 10, the vertical direction of the image pickup area of the CCD camera 12 is changed. It is set as the scanning direction.

With this setting, if there is a large brightness difference between vertically adjacent pixels in the image data of the CCD camera 12, that point is captured as an edge point. Therefore, if the preceding vehicle exists in the area to be scanned, the lateral edge of the preceding vehicle is detected as the effective line segment, and in step 214 shown in FIG. 9, for example, as shown in FIG. The edge image data including a plurality of effective line segments extending in the width direction is recorded in the main RAM 24.

When the above processing is completed, the main routine shown in FIG. 8 proceeds to step 110. Step 11
In 0, the main RAM 2
From the edge image data stored in No. 4, processing for extracting the intersection point position of the vertical effective line segment and the horizontal effective line segment is performed.

As described above, the effective line segment in the vertical direction is a line segment extracted assuming that it is the vertical edge of the preceding vehicle. Further, the effective line segment in the horizontal direction is a line segment extracted assuming that it is the horizontal edge of the preceding vehicle. Therefore, those intersections can be considered to be the corners of the vehicle in rear view.

On the other hand, in a rear view of the vehicle, two end portions near the ground clearance of the vehicle, two end portions at the roof position of the vehicle, two end portions at the trunk roof position, and two end portions at the bumper position. Etc. are detected as corners. Therefore, if there is a preceding vehicle ahead of the own vehicle, it is considered that at least four intersections are detected in step 110. Moreover, since the corners of the preceding vehicle are detected symmetrically, a rectangular shape should be able to be extracted from the four detected intersections.

Therefore, if the number of intersections extracted as the corners of the vehicle in step 110 is less than four, or even if four or more intersections are detected, a rectangle cannot be extracted from those intersections, It can be determined that the intersections detected together are not the corners of the preceding vehicle.

Therefore, in the main routine shown in FIG.
After the process of step 110, the process proceeds to step 112, and it is determined whether or not each of the extracted intersections is the corner of the preceding vehicle, based on the number and arrangement of the intersections. The process of step 110 is specifically executed by the subroutine shown in FIG.

When the routine shown in FIG. 11 is started, first, at step 400, it is judged if four or more intersections have been extracted at step 110. As a result, when it is determined that four or more intersections are detected, the process of step 402 is executed.

At step 402, it is judged if a rectangle can be extracted from the extracted intersection. As a result, when the rectangle is extracted, it is determined that the plurality of intersections extracted in step 110 correspond to the corners of the preceding vehicle, the process proceeds to step 404, and it is determined that the detected object is the preceding vehicle. After this, the present routine ends.

On the other hand, if it is determined in the above step 400 that the number of the extracted intersections is less than 4, or if it is determined in the step 402 that the rectangle cannot be extracted from the intersections, the process proceeds to step 406. After determining that the detected object is not the preceding vehicle, this routine ends. When the subroutine shown in FIG. 11 is completed in this way, the main routine shown in FIG. 8 is then completed. After that, the processing from step 100 onward described above is repeatedly executed at predetermined time intervals.

As described above, the object recognition device of this embodiment is
An edge area image obtained by extracting an edge point from the imaged data captured from the CCD camera 12 and expanding the edge point in units of small areas divided on a grid (see FIGS. 6, 7, and 14 above). ~ Image shown in FIG. 18)
It recognizes the preceding vehicle. Therefore, according to the object recognition device of the present embodiment, the contour of the preceding vehicle can be clearly captured, and the accuracy of the determination regarding the presence or absence of the preceding vehicle, the detection accuracy of the corner points of the preceding vehicle, etc. are always maintained in good condition. be able to.

In the above-described embodiment, the CCD camera 12 corresponds to the above-mentioned image sensor. Also,
The DSP 16 executes the processing of the above step 206 so that the above-mentioned area dividing means performs the above step 104.
And 108, the edge point detecting means described above by executing the processing of steps 200 to 204, and the edge region setting means described above by executing the processing of steps 208 to 212 in steps 104 and 108, By executing the processing of step 110, the above-mentioned corner point specifying means is realized.

By the way, in the above-mentioned embodiment, the image pickup area of the image sensor is divided into the grid-like small areas in the process of signal processing, but the present invention is not limited to this, and the image sensor is not limited to this. It is also possible to preliminarily divide the imaging region of No. 2 into small regions on the hardware.

[0064]

As described above, according to the present invention, the vertical edge points and the horizontal edge points detected in the image pickup data can be expanded within the width of the small area. In this case, the vertical edge points corresponding to the outline of the symmetrical object for recognition,
Even in a situation where the horizontal edge points are detected separately from each other, it is possible to form a continuous vertical edge area group and horizontal edge area group corresponding to the contour of the object. Therefore, object recognition can be performed with high accuracy based on the vertical edge region group and the horizontal edge region group.

Further, according to the present invention, even in a portion where a plurality of vertical edge points are detected in close proximity and in a portion where a plurality of horizontal edge points are detected in close proximity, they are detected in the same small area. As far as possible, the widths of the vertical edge region and the horizontal edge region are not expanded. Therefore, the vertical edge region group and the horizontal edge region group having an unnecessarily wide width are not formed, and the corners of the symmetric object are accurately recognized from the intersection of the vertical edge region group and the horizontal edge region group. Can be detected.

[Brief description of the drawings]

FIG. 1 is a block configuration diagram of an object recognition device that is an embodiment of the present invention.

FIG. 2 is an example of image pickup data of a CCD camera included in the object recognition device of the present embodiment.

FIG. 3 is an example of a vertical edge screen in which vertical edges of a recognition target are extracted based on image pickup data of a CCD camera.

FIG. 4 is an example of an image pickup area of a CCD camera divided into small areas by being divided into a lattice shape.

FIG. 5 is an example of a horizontal edge screen in which a horizontal edge of a recognition target is extracted based on image pickup data of a CCD camera.

FIG. 6 is an example of a vertical edge area screen formed by expanding vertical edge points included in the vertical edge screen in small area units.

FIG. 7 is an example of a horizontal edge area screen formed by expanding the horizontal edge points included in the horizontal edge screen in small area units.

FIG. 8 is an example of a flowchart of a main routine executed in the object recognition device of this embodiment.

FIG. 9 is an example of a flowchart of a first subroutine executed in the object recognition device of the present embodiment.

FIG. 10 is an example of a flowchart of a second subroutine executed in the object recognition device of this embodiment.

FIG. 11 is an example of a flowchart of a third subroutine executed in the object recognition device of this embodiment.

FIG. 12 is an example of an edge point screen formed when the object recognition device of the present embodiment extracts a diagonal line segment that is inclined upward to the right.

FIG. 13 is an example of an edge point screen formed when the object recognition device of the present embodiment extracts a diagonal line segment having a downward slope.

FIG. 14 is an example of an edge area screen formed when the object recognition apparatus of the present embodiment extracts diagonal line segments.

FIG. 15 is an edge region screen showing a scanning region when the object recognition device of the present embodiment extracts a vertical line segment.

FIG. 16 is an edge region screen showing a scanning direction when the object recognition device of the present embodiment extracts a vertical line segment.

FIG. 17 is an edge area screen formed when the object recognition device of this embodiment extracts a vertical line segment.

FIG. 18 is an edge area screen formed when the object recognition device of the present embodiment extracts a horizontal line segment.

[Explanation of symbols]

 10 radar 12 CCD camera 12-n, 12-k, 12-k + 1, 12-k + 2 small area 16 digital signal processor (DSP) 22 input video RAM 24 main RAM 36 vehicle controller 40 preceding vehicle 40V vertical Edge straight line 40H Horizontal edge straight line 40Vw Vertical edge area 40Hw Horizontal edge area

Claims (1)

[Claims] 1. An object recognizing device for recognizing an object using an image sensor, comprising: an area dividing unit for dividing an image pickup area of the image sensor into a lattice shape in units of small areas including a predetermined number of pixels; and image pickup data. Edge point detection means for detecting a vertical edge point and a horizontal edge point therein, an edge area setting in which a small area including the vertical edge point is a vertical edge area, and a small area including the horizontal edge point is a horizontal edge area Means, and a corner point specifying means for specifying a corner point of the object based on a region where a vertical edge area group continuous in the vertical direction and a horizontal edge area group continuous in the horizontal direction intersect, Object recognition device.