JP2018048949A - Object recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an object recognition device with which it is possible to reduce the possibility of a wall-shaped object installed along a road being misrecognized as a mobile object.SOLUTION: The object recognition device acquires the distance measurement data of an object present ahead of the vehicle (S2), clusters the acquired distance measurement data (S3), and calculates an inclination of a distance measurement point group turned into a target by clustering against the traveling direction of the vehicle (S4). The object recognition device recognizes a road boundary by performing a white line recognition process on a captured image of surroundings ahead of the vehicle (S1), and calculates an inclination of the recognized road boundary against the traveling direction of the vehicle (S5). The object recognition device compares the inclination of the distance measurement point group and the inclination of the road boundary (S6), and determines, when a difference between the two is within a preset threshold, that the distance measurement point group turned into the target is a wall-shaped object installed along the road (S7).SELECTED DRAWING: Figure 8

Description

  The present invention relates to an object identification device suitable for use in an automatic driving system.

  In Patent Document 1, if the inclination of the distance measuring point group constituting the cluster with respect to the traveling direction of the host vehicle is within a preset threshold value, the target represented by the cluster is not a moving object but is installed along the road. Disclosed is a method for determining that the object is a wall-like object.

JP 2013-36978 A

  In a curved doorway with a large curvature of the road, the inclination of the wall-like object installed along the road with respect to the traveling direction of the host vehicle becomes large. In this case, in the above method, since the inclination of the distance measuring point group with respect to the traveling direction of the host vehicle exceeds the threshold value, there is a possibility that it is erroneously determined that the object is not a wall-like object even though it is a wall-like object. is there.

  The present invention has been made in view of the above-described problems, and an object of the present invention is to provide an object identification device that can reduce misidentification of a wall-like object installed along a road as a moving object. And

In order to achieve the above object, an object identification device according to the present invention includes:
Ranging data acquisition means for acquiring ranging data of an object existing in front of the host vehicle;
Clustering means for clustering the distance measurement data acquired by the distance measurement data acquisition means;
First geometric feature amount calculating means for calculating a geometric feature amount of the distance measuring point group targeted by the clustering means;
Based on the map information of the road on which the host vehicle is traveling and the position information of the host vehicle, a road boundary in front of the host vehicle is recognized, or a white line for a photographed image obtained by shooting the front of the host vehicle Road boundary recognition means for recognizing the road boundary by recognition processing;
Second geometric feature amount calculating means for calculating a geometric feature amount of the road boundary recognized by the road boundary recognition means;
A threshold in which a difference between the geometric feature amount calculated by the first geometric feature amount calculating means and the geometric feature amount calculated by the second geometric feature amount calculating means is set in advance. A determination unit that determines that the range-finding point group targeted by the clustering unit is a wall-like object installed along the road,
It is characterized by providing.

  The first geometric feature amount calculation means calculates the geometric feature amount of the distance measurement point group when the target size of the distance measurement point group targeted by the clustering means is less than a threshold value. It may be configured to be excluded from the target. According to such a configuration, even when a long vehicle such as a truck or a bus enters the road, it is possible to reduce misidentification of the long vehicle as a wall-like object.

  The determination means includes a difference between the geometric feature quantity calculated by the first geometric feature quantity calculation means and the geometric feature quantity calculated by the second geometric feature quantity calculation means. Even when the distance measuring point group targeted by the clustering means is not moving at a speed higher than a predetermined speed for a predetermined time along the shape of the road boundary, The distance measurement point group may be configured to determine that the object is a wall-like object installed along the road. According to such a configuration, it is possible to further reduce misidentification of a wall-like object as a moving object.

  The clustering means estimates the position of the distance measurement point group in the current frame based on the position and speed of the distance measurement point group targeted in the previous frame, and determines the height of the distance measurement point group in the previous frame. If the distance is less than or equal to the threshold, the distance measurement data included in the range in which the distance measurement point group is estimated to be located in the current frame may be excluded from the clustering target. According to such a configuration, a stationary object that is not a wall-like object such as a curb can be excluded from identification targets, and the amount of calculation required for object identification can be reduced.

  According to the object identification device of the present invention, the geometric feature value of the target ranging point group is compared with the geometric feature value of the road boundary, so that the road curvature is the result of object identification. Can be suppressed. In other words, according to the present invention, it is possible to correctly identify the wall-shaped object installed along the road as a wall-like object even at a curved doorway where the curvature of the road is large. Misidentification as a moving object can be reduced.

1 is a block diagram illustrating a configuration of an automatic driving system according to a first embodiment. It is a figure explaining the function of a ranging data acquisition part. It is a figure explaining the function of a clustering part. It is a figure explaining the function of the 1st geometric feature-value calculation part. It is a figure explaining the function of a road boundary recognition part. It is a figure explaining the function of the 2nd geometric feature-value calculation part. It is a figure explaining the function of a judgment part. 3 is a flowchart showing a flow of object identification processing according to the first embodiment. It is a figure explaining the function of the road boundary recognition part which concerns on Embodiment 2. FIG. 6 is a flowchart showing a flow of object identification processing according to the second embodiment. It is a figure explaining the method of preventing misidentification with a wall-like object and a elongate vehicle. 12 is a flowchart showing a flow of object identification processing according to the third embodiment. 14 is a flowchart showing a flow of object identification processing according to the fourth embodiment.

  Embodiments of the present invention will be described below with reference to the drawings. However, in the embodiment shown below, when referring to the number of each element, quantity, quantity, range, etc., unless otherwise specified or clearly specified in principle, the reference However, the present invention is not limited to these numbers. Further, the structures described in the embodiments described below are not necessarily essential to the present invention unless otherwise specified or clearly specified in principle.

Embodiment 1
FIG. 1 is a diagram illustrating a configuration of an automatic driving system according to the first embodiment. The automatic driving system performs automatic driving of a vehicle on which the automatic driving system is mounted. The automatic driving system 1 includes a control device 10 that functions as an object identification device, a lidar (LIDAR: Laser Imaging Detection and Ranging) 2 that is an autonomous recognition sensor, a radar 3, and a camera 4. These autonomous recognition sensors are connected to the control device 10 directly or via a communication network such as a CAN (Controller Area Network) built in the vehicle. The control device 10 can grasp the situation of substantially the entire periphery of the vehicle from the object information obtained by the autonomous recognition sensor.

  The automatic driving system 1 further includes a GPS (Global Positioning System) receiver 5, a map database 6, a navigation system 7, an HMI (Human Machine Interface) 8, and an actuator 9. The GPS receiver 5 is means for acquiring position information indicating the position of the host vehicle based on a signal transmitted from a GPS satellite. The map database 6 is formed in a storage such as an HDD or an SSD mounted on the vehicle, for example. The map information included in the map database 6 includes road position information, road shape information, intersection and branch point position information, road lane information, and the like. The navigation system 7 calculates the route traveled by the host vehicle based on the position information of the host vehicle measured by the GPS receiver 5 and the map information in the map database 6, and the calculated route information is transmitted via the HMI 8. While transmitting to a driver | operator, it outputs to the control apparatus 10. The HMI 8 is an interface for outputting and inputting information between the occupant and the control device 10. The actuator 9 is a device that operates according to an operation signal from the control device 10 and changes the traveling state of the vehicle by the operation. The actuator 9 is provided in each of a drive system, a braking system, and a steering system, for example. In addition to these, the automatic driving system 1 is provided with an internal sensor for obtaining information on the traveling state of the host vehicle, such as a vehicle speed sensor and an acceleration sensor.

  The control device 10 is an ECU (Electronic Control Unit) having at least one CPU, at least one ROM, and at least one RAM. The ROM stores various types of data including various programs and maps for automatic operation. Various functions are realized in the control device 10 by loading a program stored in the ROM into the RAM and executing it by the CPU. In addition, the control apparatus 10 may be comprised from several ECU.

  In FIG. 1, among functions for automatic driving that the control device 10 has, in particular, functions related to object identification are represented by blocks. The illustration of other functions for automatic operation of the control device 10 is omitted.

  The control device 10 has a function of recognizing an object existing around the host vehicle and discriminating between a wall-like object installed along the road and a moving object. This function includes a distance measurement data acquisition unit 11, a clustering unit 12, a first geometric feature amount calculation unit 13, a road boundary recognition unit 14, a second geometric feature amount calculation unit 15 provided in the control device 10, And the determination part 16 implement | achieves. However, these units 11, 12, 13, 14, 15, and 16 do not exist as hardware in the control device 10, but are realized by software when a program stored in the ROM is executed by the CPU. Is done.

  The ranging data acquisition unit 11 acquires ranging data of an object existing around the host vehicle from the autonomous recognition sensor. The distance measurement data includes information regarding the distance to the distance measurement point, information regarding the height of the distance measurement point, and information regarding the direction of the distance measurement point. The ranging data acquisition unit 11 uses information obtained by the rider 2, uses information obtained by the radar 3, uses information obtained by the camera 4, and obtains information on a plurality of types of autonomous recognition sensors by sensor fusion. The distance measurement data can be obtained by at least one of the methods used in combination. The distance measurement data acquisition unit 11 specifies the position of each distance measurement point in the travel plane based on the acquired distance measurement data. FIG. 2 shows an example of the distribution of distance measuring points on a plane (hereinafter referred to as the XY plane) in which the distance in the forward direction relative to the host vehicle is taken as the X axis and the lateral position is taken as the Y axis. .

  The clustering unit 12 clusters the ranging data acquired by the ranging data acquisition unit 11. Clustering is performed based on the position and height of each ranging point on the X plane. Further, it is performed so as to maintain continuity with the result of clustering in the previous frame. The specific method of clustering by the clustering unit 12 is not particularly limited. A known clustering method can be used. FIG. 3 shows an example of a result obtained by converting the ranging points distributed on the XY plane into targets by clustering. A rectangular frame indicates a target, and is a distance measuring point group in which the distance measuring point group surrounded by the rectangular frame is converted into a target.

  The first geometric feature amount calculation unit 13 calculates the geometric feature amount of the distance measuring point group targeted by the clustering unit 12. The geometric feature amount includes the slope of the straight line with respect to the traveling direction of the host vehicle when the target ranging point group is approximated by a straight line using a fitting method such as the least square method. Further, the geometric feature amount includes a curvature of a curve or a curvature change rate when a target distance measuring point group is approximated by a curve using a fitting method such as a least square method. The first geometric feature quantity calculation unit 13 calculates at least one of inclination, curvature, and curvature change rate. In FIG. 4, a straight line approximating the target ranging point group is drawn with a dotted line. In the example shown in FIG. 4, the inclination of the distance measuring point group with respect to the X axis, specifically, the angle formed by the straight line approximating the distance measuring point group with respect to the X axis is calculated as the first geometric feature amount. Is done.

  The road boundary recognition unit 14 recognizes the road boundary by white line recognition processing for a captured image obtained by photographing the front of the host vehicle. The white line recognition process is performed in the order of edge point detection from the image of the camera 4, edge line segment detection, edge line segment pairing, and white line boundary selection. FIG. 5 shows a method for detecting an edge point from the image of the camera 4. The image of the camera 4 is scanned in the horizontal direction from left to right, and an edge point where the pixel changes from black to white and an edge point where the pixel changes from white to black are searched. By performing this process from top to bottom in the vertical direction, an edge line segment in which edge points are arranged on a straight line is detected. If the edge line is a white line, a set of two edge lines corresponding to both edges of the white line is obtained. The pairing of edge line segments is performed in order to select edge line segments that may be white lines by excluding edge line segments that are not white line segments. The position of the edge line segment selected by pairing is compared with the past white line detection position. When there is continuity between the two, the selected edge line segment is recognized as a white line, that is, a road boundary.

  The second geometric feature amount calculation unit 15 maps the road boundary recognized by the road boundary recognition unit 14 to the XY plane, and calculates the geometric feature amount of the road boundary on the XY plane. The geometric feature amount calculated by the road boundary recognition unit 14 includes the slope of the straight line with respect to the traveling direction of the host vehicle when the point boundary data of the road boundary is approximated by a straight line using a fitting method such as the least square method. And curve curvature or curvature change rate when the road boundary point sequence data is approximated by a curve using a fitting method such as the least square method. The second geometric feature amount calculation unit 15 is the same geometric feature as the geometric feature amount calculated by the first geometric feature amount calculation unit 13 among the slope, curvature, and curvature change rate. Calculate the quantity. The calculation of the geometric feature amount by the second geometric feature amount calculation unit 15 is performed on the road boundary in the frame indicating the target set by the clustering unit 12. In FIG. 6, a straight line approximating the road boundary in the target is drawn as a solid line. In the example shown in FIG. 6, the inclination of the road boundary with respect to the X axis, specifically, the angle formed by the straight line approximating the road boundary with respect to the X axis is calculated as the second geometric feature amount.

  The determination unit 16 includes a first geometric feature amount calculated by the first geometric feature amount calculation unit 13 and a second geometric feature amount calculated by the second geometric feature amount calculation unit 15. The difference from the target feature is calculated for each target. For example, when the inclination and the curvature are calculated as the geometric feature amount, the determination unit 16 calculates a difference in inclination and a difference in curvature, respectively. In FIG. 7, a straight line that approximates the target ranging point group is drawn with a dotted line, and a straight line that approximates a road boundary in the target is drawn with a solid line. In the example shown in FIG. 7, the difference between the inclination of the distance measuring point group and the inclination of the road boundary with respect to the X axis, specifically, the angle formed by the straight line approximating the distance measuring point group and the straight line approximating the road boundary. Calculated.

  When there is no difference between the first geometric feature quantity and the second geometric feature quantity, or when the difference is very small, the ranging point group targeted by the clustering unit 12 is There is a high possibility that the object is a wall-like object installed along the road. When the absolute value of the difference is within a preset threshold, the determination unit 16 determines that the target ranging point group is a wall-like object installed along the road. Based on the determination result, the determination unit 16 determines whether to include the target ranging point group as a tracking processing target or exclude it from the tracking processing target.

  The control device 10 functions as an object identification device by the functions of the above-described units 11, 12, 13, 14, 15, and 16. FIG. 8 is a flowchart showing the flow of object identification processing according to the present embodiment. The control device 10 as the object identification device repeatedly performs the processing shown in this flowchart at a predetermined cycle.

  First, in step S1, a road boundary in front of the host vehicle is recognized by white line recognition processing on the image of the camera 4. In parallel with the process of step S1, the processes of steps S2 and S3 are performed. In step S2, distance measurement data of an object existing around the host vehicle is acquired by the rider 2 or the radar 3. In step S3, the distance measurement data acquired in step S2 is clustered, and the distance measurement point group is converted into a target.

  In step S4, for each target, the geometric feature amount of the distance measuring point group that has been targeted in step S3 is calculated. Here, it is assumed that the gradient of the distance measurement point group is calculated as the geometric feature amount of the distance measurement point group. In step S5, the geometric feature amount of the road boundary recognized in step S1 is calculated for each target. The geometric feature amount calculated here is the same as the geometric feature amount calculated in step S4.

  In step S6, the slope of the distance measuring point group calculated in step S4 is compared with the slope of the road boundary calculated in step S5. Then, the absolute value of these differences is calculated. In step S7, it is determined whether or not the difference (absolute value) between the inclination of the distance measuring point group calculated in step S6 and the inclination of the road boundary is within a threshold value. If the difference in inclination is within the threshold, step S8 is selected as the next process, and if the difference in inclination is greater than the threshold, step S9 is selected as the next process.

  In step S8, the distance measuring point group targeted in step S3 is excluded from the tracking process. This is because, when there is no clear difference between the inclination of the distance measurement point group and the inclination of the road boundary, the distance measurement point group is highly likely to be a wall-like object installed along the road. On the other hand, in step S9, the distance measuring point group that is the target in step S3 is the target of the tracking process. Then, target information related to the distance measuring point group is output. The target information includes information such as the position, speed, and acceleration of the target. The target information is used for vehicle operation control such as inter-vehicle distance control and lane change control.

  According to the object identification process performed in the above procedure, the geometric feature value of the target ranging point group is compared with the geometric feature value of the road boundary. It can suppress affecting the identification result. Therefore, according to the automatic driving system according to the present embodiment, it is possible to correctly identify the identification accuracy of the wall-like object installed along the road as a wall-like object even at a curved doorway where the curvature of the road is large. It is possible to reduce misidentification of a wall-like object as a moving object.

Embodiment 2
Similar to the automatic driving system according to the first embodiment, the automatic driving system according to the present embodiment has the configuration shown in FIG. The difference between the automatic driving system according to the present embodiment and the automatic driving system according to the first embodiment is in the road boundary recognition method by the road boundary recognition unit 14.

  In the present embodiment, the road boundary recognition unit 14 recognizes the road boundary by self-position estimation using the image of the camera 4 and map information. FIG. 9 shows a self-position estimation method. First, the image of the camera 4 is projected on a plane so that the position of the host vehicle is the center. Further, a map image around the host vehicle is searched from the map database 6 based on the position of the host vehicle specified by the GPS receiver 5. Then, the camera image projected on the plane and the map image are collated. Since the approximate position of the road boundary is known from the map image, the point sequence data of the road boundary can be extracted from the camera image by comparing the two images.

  FIG. 10 is a flowchart showing the flow of object identification processing according to the present embodiment. In the object identification process according to the present embodiment, the process of step S1 ′ is performed instead of the process of step S1 of the first embodiment. In step S1 ′, the road boundary is recognized by self-position estimation using the image of the camera 4 and map information. The remaining processes from step S2 to step S10 are not changed from the first embodiment. As a road boundary recognition method, it is possible to use both road boundary recognition by self-position estimation using the image of the camera 4 and map information and road boundary recognition by white line recognition processing.

Embodiment 3
Similar to the automatic driving system according to the first embodiment, the automatic driving system according to the present embodiment has the configuration shown in FIG. In the automatic driving system according to the present embodiment, new functions for preventing misidentification of a long vehicle such as a truck or a bus and a wall-like object are the first geometric feature quantity calculation unit 13 and the determination unit 16. And given to.

  In the present embodiment, the first geometric feature quantity calculation unit 13 calculates the target size of the distance measuring point group that has been targeted by the clustering unit 12. The target size is at least one of the width, length, and ratio of the rectangular frame indicating the target. For the target size, a threshold value for distinguishing between a wall-like object and a long vehicle is set. When the target size is less than the threshold value, the first geometric feature amount calculation unit 13 excludes the distance measurement point group indicated by the target from the target for calculating the geometric feature amount. There is a high possibility that the excluded ranging point group is a moving object that is not a wall-like object. Therefore, the determination unit 16 includes the distance measurement point group excluded by the first geometric feature value calculation unit 13 as a target of tracking processing, and outputs target information related to the distance measurement point group.

  Further, in the present embodiment, the determination unit 16 determines whether the target distance measuring point group is moving along the shape of the road boundary for a predetermined time at a speed equal to or higher than a predetermined speed. And wall-like objects. Specifically, the determination unit 16 determines that the speed equal to or higher than a predetermined threshold has continued for a predetermined time (Condition A) and that the lateral position error of the target position at each time with respect to the lane center position continues for a predetermined time or longer. Whether or not both (condition B) are satisfied. In the example shown in FIG. 11, the speed of the target ranging point group is calculated from the change in the position of the target at times T-2, T-1, and T, and is compared with the threshold speed. A lateral position error of the target position with respect to the center position is calculated. The lane center position is calculated from the positions of the left and right white lines. If either condition A or condition B is not satisfied, the determination unit 16 determines that the distance measurement point group is likely to be a long vehicle, and the target information related to the distance measurement point group Cancel the output of.

  FIG. 12 is a flowchart showing the flow of object identification processing according to the present embodiment. In the object identification process according to the present embodiment, the process of step S101 is performed in addition to the object identification process according to the first or second embodiment. The process of step S101 is performed after the process of step S3 and before the process of step S4. In step S101, it is determined whether or not the target size of the distance measuring point group targeted in step S3 is greater than or equal to a threshold value. When the target size is equal to or larger than the threshold value, the process proceeds to step S4, and the geometric feature amount (for example, inclination) of the distance measuring point group is calculated. However, if the target size is less than the threshold value, steps S4 to S7 are skipped and the process proceeds to step S9, and the distance measuring point group targeted in step S3 is the target of the tracking process. By adding the processing in step S101, it is possible to prevent a long vehicle such as a truck or a bus from being mistaken as a wall-like object.

  Furthermore, in the object identification process according to the present embodiment, the process of step S102 is performed in addition to the object identification process according to the first or second embodiment. The process of step S102 is performed after the process of step S9 and before the process of step S10. In step S102, the past movement history of the target to be tracked is analyzed using the white line information obtained in step S1. And if both the above-mentioned conditions A and B are satisfied, it will progress to Step S10 and will output target information. However, if either of the above conditions A and B is not satisfied, the output of the target information is stopped. By performing the process of step S102, it is possible to suppress the misidentification of the wall-like object as a long vehicle such as a truck or a bus.

Embodiment 4
Similar to the automatic driving system according to the first embodiment, the automatic driving system according to the present embodiment has the configuration shown in FIG. The difference between the automatic driving system according to the present embodiment and the automatic driving system according to the first embodiment is in the function of reducing the amount of calculation applied to the object identification given to the clustering unit 12.

  In the present embodiment, as a method of reducing the amount of calculation required for object identification, clustering is performed by thinning out distance measuring points. There are the following two methods for thinning the distance measuring points. First, as a procedure common to the first method and the second method, the position of the target in the current frame is estimated using the position and speed of the distance measuring point group that was targeted in the previous frame. . Based on the assumption that the object is a rigid body, a target frame having the same height, width, and length as the previous frame is used as the target frame in the current frame.

  In the first method, the distance measurement points that fall within the target frame in the current frame are thinned at fixed intervals, or thinned by changing the thinning amount according to the distance size (for example, the distance is The closer the closer, the larger the amount to be thinned out). In the second method, when the size of the target frame in the previous frame, particularly the height, is equal to or less than the threshold value, all ranging points that fall within the target frame in the current frame are excluded from the clustering targets. By implementing such a method, the amount of calculation required for object identification can be reduced. In particular, by performing the second method, a stationary object that is not a wall-like object such as a curb can be excluded from the identification target, and the amount of calculation required for object identification can be reduced.

  FIG. 13 is a flowchart showing the flow of object identification processing according to the present embodiment. In the object identification process according to the present embodiment, the process of step S201 is performed in addition to the object identification process according to the first or second embodiment. The process of step S201 is performed after the process of step S2 and before the process of step S3. In step S201, the distance measuring points are thinned out from the distance measuring points acquired in step S2 by any one of the first method and the second method. In step S3, clustering is performed on the thinned distance measuring points. Since the content after step S4 is not changed from the first embodiment or the second embodiment, the description in the flowchart is omitted.

1 Automatic Driving System 2 Rider 3 Radar 4 Camera 5 GPS Receiver 6 Map Database 7 Navigation System 8 HMI
9 Actuator 10 Control device (object identification device)
11 Ranging data acquisition unit (Ranging data acquisition means)
12 Clustering unit (clustering means)
13 1st geometric feature-value calculation part (1st geometric feature-value calculation means)
14 Road boundary recognition part (road boundary recognition means)
15 Second geometric feature amount calculation unit (second geometric feature amount calculation means)
16 Judgment part (judgment means)

Claims (1)

Ranging data acquisition means for acquiring ranging data of an object existing in front of the host vehicle;
Clustering means for clustering the distance measurement data acquired by the distance measurement data acquisition means;
First geometric feature amount calculating means for calculating a geometric feature amount of the distance measuring point group targeted by the clustering means;
Based on the map information of the road on which the host vehicle is traveling and the position information of the host vehicle, a road boundary in front of the host vehicle is recognized, or a white line for a photographed image obtained by shooting the front of the host vehicle Road boundary recognition means for recognizing the road boundary by recognition processing;
Second geometric feature amount calculating means for calculating a geometric feature amount of the road boundary recognized by the road boundary recognition means;
A threshold in which a difference between the geometric feature amount calculated by the first geometric feature amount calculating means and the geometric feature amount calculated by the second geometric feature amount calculating means is set in advance. A determination unit that determines that the range-finding point group targeted by the clustering unit is a wall-like object installed along the road,
An object identification device comprising:

Images