JP7365286B2 - Road shape recognition device and road shape recognition method - Google Patents

Description

The present disclosure relates to a technology for recognizing road shape.

When recognizing a road shape, the road shape is estimated using information indicating the road shape, such as lane markers (for example, Patent Document 1).

Japanese Patent Application Publication No. 2002-109695

Recognized road shapes are used, for example, in driving support control and automatic driving control, and in order to improve the accuracy of these controls, recognition of road shapes over longer distances is required. On the other hand, in the conventional technology, there is a problem that the recognition accuracy of the road shape decreases as the distance over which the road shape must be recognized becomes long.

Therefore, even when the road shape recognition distance is long, there is a need to prevent or suppress a decrease in road shape recognition accuracy.

The present disclosure can be implemented as the following aspects.

A first aspect provides a road shape recognition device. The road shape recognition device according to the first aspect includes a feature point acquisition unit that obtains a plurality of feature points indicating the road shape at a predetermined timing, and a road shape recognition distance that is determined by a predetermined classification distance. Divide into a plurality of sections, generate a section road shape line for each section using current feature points acquired at the current timing and past feature points acquired at past timing, and generate the section road shape line. a road shape determining unit that determines a road shape by connecting the current feature point and the past feature point, and when the current feature point position and the past feature point position are within a predetermined distance range, the current feature point position and a road shape determination unit that generates the section road shape line using an average feature point obtained by averaging the positions of the past feature points and the positions of the past feature points.

According to the road shape recognition device according to the first aspect, even when the road shape recognition distance is long, it is possible to prevent or suppress a decrease in road shape recognition accuracy.

A second aspect provides a road shape recognition device. The road shape recognition device according to the second aspect includes a feature point acquisition unit that acquires a plurality of feature points indicating a road shape at a predetermined timing, and a feature point acquisition unit that acquires a plurality of feature points indicating a road shape at a predetermined timing. The recognition distance is divided into a plurality of sections by the segment distance, a section road shape line is generated for each section using the acquired feature points, and the road shape is determined by connecting the generated section road shape lines. a road shape determining unit that selects the segment distance that reduces the error of each of the determined road shapes, and selects a road shape that corresponds to the selected segment distance; Be prepared.

According to the road shape recognition device according to the second aspect, even when the road shape recognition distance is long, it is possible to prevent or suppress a decrease in road shape recognition accuracy.

A third aspect provides a road shape recognition device. A road shape recognition device according to a third aspect includes a feature point acquisition unit that obtains a plurality of feature points indicating a road shape at a predetermined timing, and a road shape recognition distance that is determined by a predetermined classification distance. A road shape determination unit that divides the road into a plurality of sections and generates a section road shape line for each section using the characteristic points, and determines the road shape by connecting the section road shape lines so that they are continuous. and a road shape determination unit.

According to the road shape recognition device according to the third aspect, even if the road shape recognition distance is long, it is possible to prevent or suppress a decrease in road shape recognition accuracy.

A fourth aspect provides a road shape recognition method. The road shape recognition method according to the fourth aspect acquires a plurality of feature points indicating the road shape at a predetermined timing, and divides the recognition distance of the road shape into a plurality of sections at predetermined segmentation distances. When generating a section road shape line for each section using the current feature point acquired at the current timing and the past feature point acquired at the past timing, the position of the current feature point and the past feature If the position of the point is within a predetermined distance range, the section road shape line is generated using an average feature point obtained by averaging the current feature point position and the past feature point position. and determining a road shape by connecting the section road shape lines.

According to the road shape recognition method according to the fourth aspect, even when the road shape recognition distance is long, it is possible to prevent or suppress a decrease in road shape recognition accuracy.

A fifth aspect provides a road shape recognition method. The road shape recognition method according to the fifth aspect acquires a plurality of feature points indicating the road shape at a predetermined timing, and calculates the recognition distance of the road shape for each of a plurality of predetermined segment distances. When dividing the road into a plurality of sections based on the segment distance, generating a section road shape line for each section using the acquired feature points, and connecting the generated section road shape lines to determine the road shape, The road shape is determined by selecting the division distance that reduces the error of each of the determined road shapes, and selecting the road shape corresponding to the selected division distance.

According to the road shape recognition method according to the fifth aspect, even when the road shape recognition distance is long, it is possible to prevent or suppress a decrease in road shape recognition accuracy.

A sixth aspect provides a road shape recognition method. The road shape recognition method according to the sixth aspect acquires a plurality of feature points indicating the road shape at a predetermined timing, and divides the recognition distance of the road shape into a plurality of sections at predetermined segmentation distances. The method further includes determining a road shape by connecting each of the section road shape lines so that they are continuous when generating a section road shape line for each section using the feature points.

According to the road shape recognition method according to the sixth aspect, even when the road shape recognition distance is long, it is possible to prevent or suppress a decrease in road shape recognition accuracy.

The present disclosure can also be realized as a road shape recognition program or a computer-readable recording medium that records the program.

FIG. 1 is an explanatory diagram showing an example of a vehicle equipped with the road shape recognition device according to the first embodiment. FIG. 1 is a block diagram showing the functional configuration of a road shape determination device according to a first embodiment. An explanatory diagram showing an example of a general correspondence relationship between feature points and road shape lines for all sections. 5 is a flowchart showing a process flow of road shape recognition processing executed by the road shape recognition device according to the first embodiment. An explanatory diagram showing an example of a point group of acquired past feature points. An explanatory diagram showing an example of a point group of acquired current feature points. An explanatory diagram showing an example of a point group in which corrected past feature points and current feature points are superimposed. FIG. 8 is an explanatory diagram showing an example of generating average feature points in the point group shown in FIG. 7; FIG. 2 is an explanatory diagram showing an example of acquiring feature points in a running vehicle. 10 is an explanatory diagram showing an example of a case where a section road shape line is generated in the example shown in FIG. 9. FIG. FIG. 3 is an explanatory diagram showing an example of a section road shape line obtained when the section is short and feature points are discrete. FIG. 3 is an explanatory diagram showing an example of a section road shape line and an entire section road shape line obtained when changing the segment distance. 7 is a flowchart showing a process flow of road shape recognition processing executed by the road shape recognition device according to the second embodiment. FIG. 3 is an explanatory diagram showing the effect of continuously connecting section road shape lines of two sections. 10 is a flowchart showing a process flow of road shape recognition processing executed by a road shape recognition device according to a third embodiment.

A road shape recognition device and a road shape recognition method according to the present disclosure will be described below based on several embodiments.

First embodiment:
As shown in FIG. 1, the road shape recognition device 10 according to the first embodiment is used, for example, by being mounted on a vehicle 50. The road shape recognition device 10 includes at least a road shape determination device 100 that determines the road shape. The road shape determination device 100 functions as a feature point acquisition unit and a road shape determination unit. The road shape determination device 100 is connected to a detector 20 for detecting objects outside the vehicle 50 via a signal line. The detector 20 includes a millimeter wave radar 21, a camera 22, and a lidar (Light Detection and Ranging) 23. Note that, hereinafter, the millimeter wave radar 21, camera 22, and lidar 23 may be collectively referred to as the detector 20. In the example of FIG. 1, the millimeter wave radar 21 and the lidar 23 are placed on the front grill and front bumper of the vehicle 50, and the camera 22 is placed on the top of the front window. The millimeter wave radar 21, the camera 22, and the lidar 23 may be placed elsewhere in the vehicle 50. For example, the millimeter wave radar 21 and the lidar 23 may be placed on the side or rear of the vehicle 50, for example, on the side or rear of the vehicle 50. The camera 22 may be placed on the front fender, roof, rear fender, or rear bumper, and the camera 22 may be placed on the side or rear of the vehicle 50, for example, on the front fender, roof, rear fender, rear bumper, rear window, or door mirror of the vehicle 50. It's okay. Objects in the target direction can be detected by the millimeter wave radar 21 and lidar 23 located at these other locations. The millimeter wave radar 21 and lidar 23 are also called distance measuring devices, and can detect not only the distance to an object, but also the position and characteristics of the object. The camera 22 acquires information about the outside world as image information. The road shape determination device 100 includes a vehicle control device 40 for controlling controlled devices such as a drive source, a steering mechanism, and a braking mechanism of the vehicle 50 in order to perform travel control, driving support control, or automatic driving control of the vehicle 50. are connected via signal lines. The vehicle 50 is further provided with various sensors (not shown) for detecting the running state of the vehicle, such as a vehicle speed sensor, a yaw rate sensor, and a steering angle sensor.

As shown in FIG. 2, the road shape determination device 100 includes a central processing unit (CPU) 101 as an arithmetic unit, a memory 102 as a storage unit, an input/output interface 103 as a feature point acquisition unit, and a clock generator (not shown). We are prepared. The CPU 101, memory 102, input/output interface 103, and clock generator are connected via an internal bus 104 so that they can communicate in both directions. The memory 102 includes a memory, such as a ROM, which is a storage unit that non-volatilely and read-only stores road shape recognition programs Pr1, Pr2, and Pr3 for recognizing road shapes, and a memory, such as a RAM, which can be read and written by the CPU 101. Contains. Note that the road shape recognition program Pr1 is used in the first embodiment, the road shape recognition program Pr2 is used in the second embodiment, and the road shape recognition program Pr3 is used in the third embodiment. The rewritable area of the memory 102 includes a feature point storage area 102a that stores detection points acquired via the detector 20 or feature points at the current or one or more past points generated using the detection points. I'm here. The CPU 101, that is, the road shape determination device 100 functions as a road shape determination unit by expanding road shape recognition programs Pr1, Pr2, and Pr3 stored in the memory 102 into a readable/writable memory and executing them. can function as part of the feature point acquisition unit together with the input/output interface 103. Note that the CPU 101 may be a single CPU, may be multiple CPUs that execute each program, or may be a multitasking type or multithreading type CPU that can simultaneously execute multiple programs. Also good.

A millimeter wave radar 21, a camera 22, and a lidar 23 are connected to the input/output interface 103 via detection signal lines. A vehicle control device 40 is connected to the input/output interface 103 for two-way communication, and vehicle information 60 indicating the running state of the vehicle is input.

The millimeter wave radar 21 is a sensor that detects the distance, relative speed, and angle of an object with respect to the vehicle 50 by emitting millimeter waves and receiving reflected waves reflected by the object. The camera 22 is an imaging device including an imaging device such as a CCD or an imaging device array, and is a sensor that receives visible light and outputs external shape information or shape information of the object as image data as a detection result. The lidar 23 is a sensor that detects the distance, relative speed, and angle of an object with respect to the vehicle 50 by emitting infrared laser light and receiving reflected light reflected by the object. In this embodiment, each detector 20 is used to detect detection points for obtaining feature points indicating the road shape. That is, when the detector 20 detects one type of detection point corresponding to one target object, the detection point is used as it is as a feature point, and the detector 20 detects multiple types of detection points corresponding to multiple targets. When detection points are detected, one feature point is acquired using a plurality of detection points, as will be described later. The feature point indicating the road shape is an object that can represent the road shape, such as a roadway center line indicating the center of the road, a lane boundary line dividing the lanes, etc., drawn on the roadway with paint, studs, stones, etc. It is acquired using detection points obtained from road markings including roadway outer lines that separate the roadway from the road shoulder, curbs placed along the roadway, protective fences such as guardrails, and wall structures such as soundproof walls and soundproof walls. That is, detection points acquired by the detector 20 corresponding to non-smooth portions on the surface of the object are used as feature points indicating the road shape. More specifically, the detection wave emitted from the millimeter wave radar 21 or the detection light emitted from the lidar 23 is reflected at, for example, a step between a marking line and a road surface, that is, an edge, and a detection point is obtained by reflecting the detection wave or a wall surface. Detection points obtained by reflection from above and detection points obtained by reflection from edges of curbs and guard fences can be used as feature points indicating the road shape. Further, in the image data captured by the camera 22, detection points obtained by discretely extracting the above-mentioned steps and edges can be used as feature points indicating the road shape. In the camera 22 and the lidar 23, steps and edge portions are detected as areas where the brightness value changes greatly. Furthermore, detection points corresponding to the travel trajectories of other vehicles traveling around the host vehicle can be used as feature points indicating the road shape. The travel trajectory of the other vehicle is obtained by acquiring over time the coordinate position of an arbitrary position in the other vehicle area, which is extracted by pattern matching or the like from detection points and image data of the other vehicle.

As shown in FIG. 3, generally, in order to smoothly transition from a straight section to a curved section, a curved section of a road and a straight section are connected via a relaxation section. The relaxation section is defined using, for example, a clothoid curve. The road shape is determined by generating a fitting curve including at least one of a straight line and a curved line, that is, a road shape line L1, generated by the least squares method using feature points DP indicating the road shape. When the road shape recognition distance is narrow or short, for example, when it is about 50 meters from the host vehicle VC, the fitting curve L1 and the feature points DP match well, and the road shape can be recognized with high accuracy using the feature points. be able to. However, when the recognition distance of the road shape exceeds about 60 m from the own vehicle VC, the fitting curve L1 and the feature points DP do not match, the recognition accuracy of the road shape decreases, and furthermore, curved sections are included. In this case, the reduction in road shape recognition accuracy becomes even greater. In order to solve this problem, the recognition distance of the road shape is divided into multiple sections by a predetermined distance, the section road shape line is determined for each section, and the obtained multiple section road shape lines are connected. One possible method is to determine and recognize the road shape using In this case, while it is expected that the degree of coincidence between the feature points DP and the fitting curve L1 in each section will improve, this is due to the fact that the road shape line is generated for each divided section. Problems also arise.

In this embodiment, the number of feature points in each section, which may occur when the recognition distance of the road to be recognized, that is, the recognition distance of the road shape, is divided into a plurality of sections with predetermined segmentation distances is reduced. It is possible to suppress or prevent a decrease in road shape recognition accuracy due to shortage or absence. The segment distance is a distance that defines the segment length that can be selected by dividing the recognition distance, and the predetermined segment distance, that is, the distance of each segment, ensures that the error of the fitting curve is within a predetermined tolerance. Maximum length is set. A road shape determination process executed by the road shape determination device 100 according to the first embodiment will be described. The processing routine shown in FIG. 4 is repeatedly executed at predetermined time intervals, for example, every few milliseconds, after the control system of the vehicle 50 is started or the start switch is turned on. The processing flow shown in FIG. 4 is executed by the CPU 101 executing the road shape recognition program Pr1.

The CPU 101 acquires a plurality of past feature points DP1 at a first timing (step S100), and acquires a plurality of current feature points DP2 at a second timing (step S102). The plurality of past feature points DP1 and the plurality of current feature points DP2 that have been acquired each form a point group or a point sequence group. Detection points, that is, feature points are acquired by the detector 20 at predetermined time intervals. The CPU 101 may acquire feature points from the detector 20 in real time, or may acquire feature points detected by the detector 20 and temporarily stored in the memory 102 at a desired timing. The first timing means a timing that is temporally past the second timing, which is the current timing, and is the acquisition timing of feature points immediately before the second timing, or a feature point that is past the timing immediately before the second timing. It may be any point acquisition timing. Note that the number of sampling times is not limited to two, and the first timing may be a plurality of timings going back in the past. By using feature points at a plurality of first timings, the number of samplings increases and the reliability of the obtained section road shape line RL improves. Note that when this processing routine is executed for the first time and thereafter, the average feature point DP3 may be used as the past feature point DP1, or the current feature point DP2 may be used as the past feature point DP1. When the average feature point DP3 is used as the past feature point DP1, unnecessary fluctuations are suppressed, the reliability of the past feature point DP1 is improved, and the accuracy of road shape recognition is also improved.

FIG. 5 illustrates the past feature point DP1 acquired at the first timing. In FIG. 5, the road shape recognition distance is divided into six sections, a first section S1 to a sixth section S6. FIG. 6 illustrates the current feature point DP2 acquired at the second timing. In FIG. 6, no feature point DP2 currently exists in the fourth section S4 and the fifth section S5. As described above, the feature points are the detection results of non-smooth parts of the object, and depending on the attitude of the vehicle 50 while running, the non-smooth parts, that is, edges and edge parts may not be detected well. As a result, all or part of the feature points may be missing in some areas of the road shape recognition distance.

The CPU 101 calculates average features for the past feature point DP1 and the current feature point DP2 that are within a predetermined distance range among the past feature point DP1 obtained at the first timing and the current feature point DP2 obtained at the second timing. Point DP3 is acquired (step S104). Note that since the vehicle VC is moving between the first timing and the second timing, the relative position of the past feature point DP1 with respect to the vehicle VC changes, that is, the vehicle VC approaches the past feature point DP1, and the vehicle VC The distance between the past feature point DP1 and the past feature point DP1 is short. Therefore, in this embodiment, as shown in FIG. 7, past feature point DP1 is corrected according to the amount of movement of vehicle VC, that is, ego motion correction is performed, and corrected past feature point DP1m is obtained. In FIG. 7, the corrected past feature point DP1m and the current feature point DP2 are shown superimposed. As shown in FIG. 8, an average feature point DP3 of the current feature point DP2 and the corrected past feature point DP1m is calculated. Note that the amount of movement of the vehicle VC may be obtained by using distance information provided for display on the odometer, and may be obtained by using the average speed of the vehicle VC between the first timing and the second timing and the first timing. It may be calculated using the time interval between the second timing and the second timing. FIG. 8 shows a corrected past feature point DP1m that was not subject to averaging, a current feature point DP2, and an average feature point DP3 calculated by averaging.

The CPU 101 generates a section road shape line RL using the corrected past feature point DP1m, current feature point DP2, and average feature point DP3 in each section (step S106). Specifically, the CPU 101 uses the least squares method to calculate a quadratic function representing a straight line or curve connecting the feature points DP1m, DP2, and DP3 existing in each section. That is, the section road shape line RL is a line that indicates the road shape for each section, which is generated using the feature points DP. In the example of FIG. 8, for the first section S1 and the third section S3, the section road shape line RL is generated using the average feature point DP3, and for the sixth section S6, the corrected past feature point DP1m and the average feature point A section road shape line RL is generated using DP3. Regarding the fourth section S4 and the fifth section S5, since the current feature point DP2 does not exist, the section road shape line is generated using the corrected past feature point DP1m. In addition, when generating the section road shape line RL, the amount of positional deviation of the current feature point DP2 with respect to the section road shape line RL generated at a past timing, preferably the immediately previous timing, is greater than or equal to a predetermined determined positional deviation amount. If so, it is desirable to delete the current feature point DP2. In the present embodiment, in the second section S2, the position of the current feature point DP2 is largely deviated from the section road shape line RL generated at a past timing, so the current feature point DP2 is Excluded from the feature points used to generate shape lines. As a result, the section road shape line RL is generated using the corrected past feature point DP1m and the average feature point DP3. In this way, the accuracy of the section road shape line RL is improved by excluding feature points detected at positions significantly different from the positions of feature points at different timings in time series. Note that the determination as to whether the current feature point DP2 is significantly different from the position of the feature point at a past timing is determined based on the section road shape line RL generated at a past timing, as described above. Alternatively, the determination may be made based on the position of the past feature point DP1 or the corrected past feature point DP1m used to generate the section road shape line RL at a past timing.

The CPU 101 connects the six section road shape lines RL generated corresponding to each section S1 to S6 to generate one road shape line, that is, the entire section road shape line, and The road shape is determined by using the road shape (step S108), and this processing routine is ended. The determined road shape is defined by a quadratic function representing the entire section road shape line, ie, a quadratic function representing each section road shape line RL with appropriate boundary processing performed. As a result, by using the quadratic function, it is possible to easily perform driving support and automatic driving control according to the road shape in the subsequent stage.

An example of how feature points are acquired in a running vehicle VC will be described with reference to FIGS. 9 and 10. Regardless of whether the detector 20 is single or multiple, or whether the detector 20 is of a single type or multiple types, FIGS. 9 and 10 show that the detector 20 detects multiple types of targets. An example is shown in which detection points indicating multiple types of road shapes are acquired from an object. In the examples of FIGS. 9 and 10, the first type detection point LP corresponds to the lane boundary line LM1, the second type detection point RP corresponds to the wall RE, and the third type detection point VP corresponds to the traveling trajectory of the other vehicle EVC. is exemplified. In addition to this, a fourth type detection point (not shown) corresponding to the outer road line or the curb LM2 may be used. A feature point DP is obtained by appropriately combining the obtained plural types of detection points LP, RP, and VP. For example, for a recognition distance close to the vehicle VC where the recognition accuracy of the lane boundary line LM1 is higher than the desired recognition accuracy, the first type detection point LP is used. Regarding the two first type detection points LP, the intersection between the line segment connecting the two first type detection points LP and the center line of the parallel lane boundary line LM1, which is the center line of the lane, is calculated, and the intersection point is calculated as the feature point DP. is obtained as. Alternatively, a point where a line segment is extended from the two first type detection points LP in a direction parallel to the width direction of the vehicle VC by the distance from the predetermined lane boundary line LM1 to the center line of the lane boundary line LM1. is acquired as the feature point DP. For recognition distances far from the vehicle VC where the angle of the optical axis of the detector 20 with respect to the road surface becomes small and the recognition accuracy of the lane boundary line LM1 decreases, the second type detection point RP and the third type detection point VP are used. It will be done. The intersection point between a line segment extending from the second type detection point RP or the third type detection point VP in a direction parallel to the width direction of the vehicle VC and the center line of the parallel lane boundary line LM1, which is the center line of the lane, is calculated. and is acquired as a feature point DP. Alternatively, the position of the travel trajectory of the other vehicle EVC is defined as the center line of the adjacent lane, and the second type detection is performed only by the distance from the predetermined wall RE or the center line of the adjacent lane to the center line of the lane boundary line LM1. A point obtained by extending a line segment from the point RP or the third type detection point VP in a direction parallel to the width direction of the vehicle VC is acquired as a feature point DP. As shown in FIG. 10, a section road shape line RL is generated for each of the plurality of divided sections using the acquired feature points DP. Note that when only one detection point is output by the detector 20, the output detection point is acquired as a feature point DP, or the output feature point is located at the center of the driving lane of the vehicle VC. Feature points are acquired by performing conversion processing.

According to the road shape recognition device 10 according to the first embodiment described above, since the road shape recognition distance is divided into a plurality of sections with predetermined segment distances, the road shape recognition distance is divided into a plurality of sections with predetermined segment distances. Deterioration in shape recognition accuracy can be suppressed or prevented. Further, according to the road shape recognition device 10 according to the first embodiment, the past feature point DP1, the current feature point DP2, and the position of the current feature point DP2 and the position of the past feature point DP1 are within a predetermined distance range. A section road shape line RL is generated for each section using the average feature point DP3 found when It is possible to suppress or prevent a decrease in the generation accuracy of section road shape lines due to missing or insufficient feature points, which may occur when road shape lines are generated for each of a plurality of sections. Furthermore, since the average feature point DP3 is generated only when the past feature point DP1 or the corrected past feature point DP1m and the current feature point DP2 are close to each other, the reliability of the average feature point DP3 is high. Since the current feature point DP2 that cannot generate the average feature point DP3 is not used, by introducing the concept of the average feature point DP3, it is possible to improve the generation accuracy of the section road shape line, which reduces the reduction in road shape recognition accuracy. It is possible to suppress or prevent this, and further improve the recognition accuracy of road shape.

According to the road shape recognition device 10 according to the first embodiment, when using the past feature point DP1, the corrected past feature point DP1m that has undergone ego motion correction is used, so that the movement of the host vehicle VC cannot be reflected. This makes it possible to further improve road shape recognition accuracy.

According to the road shape recognition device 10 according to the first embodiment, since the average feature point DP3 is used, when the whole section road shape line at the first timing and the whole section road shape line at the second timing are superimposed, , the storage capacity required for the memory 102 or the feature point storage area 102a can be reduced compared to the case where the past feature point DP1 and the current feature point DP2 are each stored in the feature point storage area 102a.

In the first embodiment, when acquiring the average feature point DP3, the reliability of multiple types of detection points is not taken into account. On the other hand, the weighted average feature point DP3 may be obtained by weighted averaging processing after considering the reliability of a plurality of types of detection points. In general, the detection accuracy of the lane boundary line LM1 is high, so for example, the first type detection point LP is assigned a reliability of 90%, and the existence probability of the wall RE is low compared to the lane boundary line LM1. In addition, since there may be cases where the shape does not follow the driving lane of the host vehicle VC, including merging and branching points, the second type detection point RP is used to detect the shape of the road in the distance. For example, the second type detection point RP is assigned a reliability of 70%, the travel trajectory of another vehicle EVC has a lower probability of existence compared to the lane boundary line LM1, and the third type detection point VP is located on a distant road. While it is used for shape detection, there is a high possibility that the other vehicle EVC will travel along a trajectory along the driving lane of the own vehicle VC, so for example, a reliability of 80% is assigned to the third type detection point VP. It will be done. Furthermore, the reliability of the ego-motion-corrected feature points is set low, for example, by about 10%, because the correction has been performed. This reliability will be used to explain the example shown in FIG. 8. For the sections S1 to S3, since the first type detection point LP is used, for example, while a reliability of 90% is assigned to the past feature point DP1, the reliability is set to 80% due to ego motion correction. Ru. A reliability of 90% is also assigned to the current feature point DP2. As a result, the reliability of the average feature point DP3 is 85%. When the average feature point DP3 is used as the next past feature point DP1, ego motion correction is applied to the reliability of 85%, for example, 75% is assigned as the reliability of the average feature point DP3. Also, the coordinates (x3, y3) of the average feature point DP3 are x3 = (80x1 + 90x2) / 170 with respect to the coordinates (x1, y1) of the corrected past feature point DP1m and the coordinates (x2, y2) of the current feature point DP2. , y3=(80y1+90y2)/170. For the section S6, since the second type detection point RP and the third type detection point VP are used, for example, 70% reliability is assigned to the past feature point DP1, and the reliability is 60% due to ego motion correction. is set to If the current feature point DP2 is obtained by the third type detection point VP, a reliability of 80% is assigned. As a result, the reliability of the average feature point DP3 is 70%. When the average feature point DP3 is used as the next past feature point DP1, ego motion correction is applied to the reliability of 70%, for example, 60% is assigned as the reliability of the average feature point DP3. Also, the coordinates (x3, y3) of the average feature point DP3 are x3 = (60x1 + 80x2) / 140 with respect to the coordinates (x1, y1) of the corrected past feature point DP1m and the coordinates (x2, y2) of the current feature point DP2. , y3=(60y1+80y2)/140. Note that among the past feature points DP1 or corrected past feature points DP1m, lower reliability is assigned to feature points DP1 and DP1m that were not used for calculating the average feature point DP3 or were used less frequently. It's okay. In this case, the weight of feature points with large errors that are not used for the average feature point DP3 is reduced, so that it is possible to suppress a decrease in accuracy of the section road shape line or to improve the accuracy.

In the first embodiment, egomotion correction may not be performed. For example, if the decrease in accuracy of the past feature point DP1 is not taken into consideration or is negligible, ego motion correction may not be performed.

Second embodiment:
When determining the road shape, the road shape recognition device according to the second embodiment prepares a plurality of predetermined segmentation distances as distances for segmenting the road shape recognition distance, and for each predetermined segmentation distance, Divide the recognition distance into multiple sections using segment distances, connect the segment road shape lines generated for each segment, and choose a segment distance from among the multiple segment distances that will reduce the error in the road shape obtained. This differs from the road shape determination device 100 according to the first embodiment in that the road shape is selected and the road shape corresponding to the selected segment distance is selected. However, the difference lies in the road shape determination process executed by the road shape recognition program Pr2, and the configuration of the road shape recognition device according to the second embodiment is different from that of the road shape recognition device 10 according to the first embodiment. Since they are the same, the same reference numerals are used to omit the explanation.

As described in the first embodiment, when the road shape recognition distance is long and one fitting curve is used, there is a strong tendency that the fitting curve and the feature points do not match, and the road shape recognition accuracy decreases. . In order to solve this problem, if the recognition distance of the road shape is divided into a plurality of sections, the following problem may occur if a fixed, predetermined division distance is used. That is, when the predetermined segment distance is long and the road shape includes a sharp curve, there is still a strong tendency that the fitting curve and the feature points do not match. On the other hand, as shown in FIG. 11, when the predetermined segment distance is short and the road shape includes a straight line or a gentle curve, false detection of feature points occurs, that is, the fitting curve for each section due to noise. That is, there is a strong tendency for the road shape line L1 for all sections to vary. In the example of FIG. 11, this can be seen in the fourth section S4 and the fifth section S5. In either case, the recognition accuracy of the road shape decreases. As shown in FIG. 12, the road shape determining device 100 according to the second embodiment prepares in advance a plurality of segmentation distances for segmenting the recognition distance according to the actual road shape to be recognized. A fitting curve, that is, a section road shape line is generated for each section, and the road shape is determined. The road shape recognition device 10 according to the second embodiment further selects a segmentation distance for segmenting that reduces the error in the road shape according to the actual road shape, and generates a section road shape line corresponding to the selected segmentation distance. The road shape determined using the following method is used.

In FIG. 12, SRoad exemplifies a case where the road includes a straight line or a gentle curve, and is divided into two sections, a first section S1 and a second section S2, for each section distance D11. The segment distance D11 is, for example, about 40 to 60 m. A section road shape line RL that approximates the feature point DP is generated corresponding to each section S1, S2, and the entire section road shape line L1 indicating the road shape is determined by connecting the generated section road shape lines RL. be done. In FIG. 12, CRoad exemplifies a case where the road includes a sharp curve, and is divided into five sections, ie, a first section S1 to a fifth section S5, for each section distance D12 that is shorter than the section distance D11. A section road shape line RL that approximates the feature point DP is generated corresponding to each section S1 to S5, and the entire section road shape line L2 indicating the road shape is determined by connecting the generated section road shape lines RL. be done. The plurality of predetermined division distances used when dividing the recognition distance are shorter when the road includes a sharp curve than when the road includes a straight line or a gentle curve.

The road shape determination process executed by the road shape recognition device 10 according to the second embodiment will be described. The processing routine shown in FIG. 13 is repeatedly executed at predetermined time intervals, for example, every few milliseconds, after the control system of the vehicle 50 is started or the start switch is turned on. The processing flow shown in FIG. 13 is executed by the CPU 101 executing the road shape recognition program Pr2. Note that the specific procedure for acquiring feature points and the procedure for generating section road shape lines using feature points have already been explained in the first embodiment, so a detailed explanation will be omitted.

The CPU 101 acquires feature points DP (step S200), and determines the road shape at the first segment distance and the second segment distance (step S202). In the examples of FIGS. 12 and 13, the predetermined segment distances, that is, the segment distances include two segment distances: a first segment distance D11 corresponding to segment distance D11 and a second segment distance D12 corresponding to segment distance D12. is used. Note that a plurality of segmentation distances different from the first and second segmentation distances may be used. The CPU 101 generates section road shape lines, which are approximate curves connecting each feature point included in each section, for each predetermined segment distance, and connects the generated section road shape lines to generate an entire section road shape line. and determine the road shape. As described above, the CPU 101 always generates the entire road shape line corresponding to the two segment distances, the first segment distance D11 and the second segment distance D12.

The CPU 101 selects the segmented distance with the smallest error from among the road shapes corresponding to the first segmented distance D11 and the second segmented distance D12 (step S204). Specifically, the CPU 101 calculates the curvature of the curved portion of the entire road shape line generated using the longer segment distance D11, and if the obtained curvature is equal to or greater than the determination threshold, It is determined that the error in the obtained road shape is larger than the error in the road shape obtained using the segment distance D12. As described above, when using the short segment distance D12, the accuracy of the road shape line for all sections is reduced due to noise related to the feature points, and therefore the accuracy of curvature determination is reduced. Therefore, the curvature is determined using the entire road shape obtained using the longer segment distance D11. If the curvature is large, it means that a sharp curve is included. Therefore, for the above-mentioned reason, the CPU 101 determines that the error in the road shape determined using the short segment distance D12 is larger than that of the long segment distance D11. It is determined that the error is smaller than the error of the road shape determined using this method. On the other hand, when the curvature is small, it means that a straight line or a gentle curve is included. Therefore, for the above-mentioned reason, the CPU 101 determines that the error of the road shape determined using the long segment distance D11 is shorter. It is determined that the error is smaller than the error in the road shape determined using the segment distance D12. The judgment threshold for determining the size of the curvature is determined by determining the upper limit of the applicable curvature once the segment distance and the fitting curve, that is, the order of the road shape line for the entire section, are determined. Set to threshold. For example, when 50 m is used as the first segment distance D11, curvature 1/R100 may be used as the determination threshold. In addition, the error in road shape is determined statistically by calculating the error between each feature point and the road shape line for the entire section, and using the average value, median value, maximum value, etc. of the calculated errors. Also good.

The CPU 101 determines as the road shape the entire section road shape line corresponding to the determined segmented distance with the small error (step S206), and ends this processing routine.

According to the road shape recognition device 10 according to the second embodiment described above, the recognition distance of the road shape is divided into a plurality of sections by the section distance for each of a plurality of predetermined section distances, and the section road shape line is created. is generated, determines the road shape, selects the segmentation distance that reduces the road shape error, and selects the road shape that corresponds to the selected segmentation distance. Decrease in recognition accuracy can be suppressed or prevented. More specifically, according to the road shape recognition device 10 according to the second embodiment, when the curvature of the road shape is less than the determination threshold, that is, when the road shape is a straight line or a gentle curve, a predetermined The entire road shape line generated using a long segment distance is determined as the road shape, and if the curvature of the road shape is equal to or greater than the determination threshold, that is, if the road shape includes a sharp curve, The entire road shape line generated using a short segment distance as a predetermined segment distance is determined as the road shape. As a result, it is possible to divide the recognition distance of the road shape into multiple sections according to the road shape, that is, with a division distance that reduces the error with respect to the road shape, and to generate a section road shape line. When dividing the recognized distance into a plurality of sections, it is possible to reduce or prevent mismatch between the divided distance and the road shape, thereby suppressing a decrease in the recognition accuracy of the road shape or improving the recognition accuracy. That is, the road shape recognition device 10 according to the second embodiment adopts a long segment distance to recognize the road when it can be determined that the road to be recognized does not include a sharp curve and the error in the road shape is small. Since the shape is determined, it is possible to suppress or prevent variations in the section road shape line RL caused by noise in the feature points DP that occurs when a short segment distance is used.

In the description of the second embodiment above, the explanation was given using an example using two segment distances, but three or more segment distances are used, and the entire interval generated according to the three or more segment distances. The segment distance may be selected according to the curvature of the road shape line, and the road shape may be determined. Furthermore, without preparing the segment distance in advance, the segment distance is dynamically set according to the curvature of the entire section road shape line generated from the reference segment distance, and the segment distance is automatically set using the set segment distance. The shape line and the road shape line for all sections may be generated, and the road shape may be determined. Dynamic determination of the segment distance can be realized, for example, by determining an appropriate segment distance for the curvature experimentally or empirically in advance, and using a map showing the correspondence between the curvature and the segment distance. .

Third embodiment:
In determining the road shape, the road shape recognition determination device according to the third embodiment generates sections for a plurality of sections obtained by dividing the road shape recognition distance by a predetermined segment distance. This embodiment differs from the road shape recognition device 10 according to the first embodiment in that when connecting the road shape lines, the road shape lines of each section are successively connected to obtain the entire road shape line. However, the difference lies in the road shape determination process executed by the road shape recognition program Pr3, and the configuration of the road shape recognition device according to the third embodiment is different from the road shape recognition device 10 according to the first embodiment. Since they are the same, the same reference numerals are used to omit the explanation.

As shown as a reference example in FIG. 14, when the recognition distance of a road shape is divided into multiple sections, when feature points DP are missing at or near the boundaries of each section, or when the number of feature points DP is small , the continuity of the section road shape lines RL1 and RL2 at the boundary may be impaired. Continuity means that two line segments are connected smoothly, and that the end points of the two straight lines match or that the connection at the end points is smooth, that is, the change in slope is small. can be defined by These events include the fact that the recognition distance is divided into multiple sections and the section road shape lines RL1 and RL2 are generated using the feature points DP that exist in each section, and that the feature points are continuous near the boundaries of the multiple sections. It occurs due to sexual unfulfillment. Therefore, when generating the section road shape lines RL1 and RL2 for each section using the feature points DP, the road shape recognition device 10 according to the third embodiment connects the section road shape lines so that they are continuous. determine the road shape. More specifically, the road shape recognition device 10 according to the third embodiment detects a road in a section far from the host vehicle VC, that is, a road in the vicinity so as to be continuous with the road shape line in the nearby road in a section close to the host vehicle VC. A distant section road shape line for a section that is farther away from the host vehicle VC than the shape line is generated. As a result, the section road shape line RL2 and the entire section road shape line RL1+RL2 shown in FIG. 14 as the third embodiment can be obtained.

A road shape determination process executed by the road shape recognition device 10 according to the third embodiment will be described. The processing routine shown in FIG. 15 is repeatedly executed at predetermined time intervals, for example, every few milliseconds, after the control system of the vehicle 50 is started or the start switch is turned on. The processing flow shown in FIG. 13 is executed by the CPU 101 executing the road shape recognition determination program Pr3. Note that the specific procedure for acquiring the feature points DP and the procedure for generating the section road shape line using the feature points DP have already been explained in the first embodiment, so a detailed explanation will be omitted.

The CPU 101 acquires feature points DP (step S300), and generates a proximate section road shape line RL1 using the feature points DP in the proximal section of the section closest to the own vehicle VC (step S302). Specifically, the CPU 101 uses the least squares method to obtain a quadratic function f(x)=a1x ² +b1x+c1 representing the adjacent road shape line RL1. The CPU 101 generates a separated section road shape line so that the separated section road shape line RL2 that follows the generated proximate section road shape line RL1 is continuously connected to the proximate section road shape line RL1 (step S304). Specifically, it is as follows. The CPU 101 defines the separated section road shape line RL2 using a quadratic function g(x)=a2x ² +b2x+c2. The continuity condition that the adjacent section road shape line RL1 and the distant section road shape line RL2 are continuous at the boundary of the section is as follows.
・Y coordinate position (horizontal position): f(x1) = g(x1),
・Slope at the connection point: f'(x1)=g'(x1),
Since the separated section road shape line RL2 passes through the farthest point EP (y2, x2),
・Farthest point passage condition: g(x2)=y2
becomes. The CPU 101 generates a quadratic function g(x)=a2x ² +b2x+c2 indicating the separated section road shape line RL2 by finding three unknown coefficients (a2, b2, c2) that satisfy these conditions.

The CPU 101 determines the road shape by generating a road shape line for all sections using the generated proximate section road shape line RL1 and distant section road shape line RL2 (step S206), and ends this processing routine.

According to the road shape recognition device 10 according to the third embodiment described above, the recognition distance of a road shape is divided into a plurality of sections with predetermined segment distances, and the section road shape is determined for each section using feature points. When generating lines, the road shape is determined by connecting the road shape lines in each section continuously, so it is possible to suppress or prevent a decrease in road shape recognition accuracy when the road shape recognition distance is long. can. More specifically, according to the road shape recognition device 10 according to the third embodiment, the adjacent section road shape line RL1 of the section close to the own vehicle VC is generated, and the own vehicle VC following the adjacent section road shape line is generated. The separated section road shape lines RL2 of the sections separated from the road shape line RL2 are continuously connected. As a result, when dividing the road shape recognition distance into multiple sections, the lack of feature points DP at or near the boundaries of each section, or the section road shape line RL1 due to lack of feature points DP, becomes a problem. The lack of continuity in RL2 can be prevented or reduced. As a result, the accuracy and reliability of the entire section road shape line obtained by connecting the section road shape lines RL1 and RL2 are improved, and the recognition accuracy of the road shape is improved. In other words, even if the feature points DP at or near the boundaries of each section are missing or feature points DP are insufficient, it is possible to suppress the decrease in accuracy of the road shape line for the entire section or to improve the accuracy. I can do it.

In the third embodiment, the case where the road shape is divided into two sections has been described, but the same applies even when the road shape is divided into three sections. That is, a section road shape line is generated in the nearest section closest to the host vehicle VC, a section road shape line for the subsequent section is generated so as to satisfy the above-mentioned continuity condition, and a section road shape line for the subsequent section is generated as described above. generated to satisfy the continuity condition. It is sufficient that this process is executed until the farthest interval is reached.

In the third embodiment, the adjacent section road shape line RL1 is generated by the least squares method, but the section road shape line to be generated by the least squares method may also be determined according to the number of feature points DP in each section. good. For example, when the number of feature points DP in the distant section is large, the distant section road shape line RL2 may be generated by the least squares method, and the adjacent section road shape line RL1 may be generated so as to satisfy the continuity condition. This is because a section road shape line of a section with a large number of feature points DP has higher reliability than a section road shape line of a section with a small number of feature points DP.

Other embodiments:
(1) Although the first embodiment, the second embodiment, and the third embodiment are each described as separate embodiments, the first to third embodiments can be executed in any combination. It's okay. In this case, multiple technical effects obtained by each embodiment can be obtained synergistically.

(2) In each of the above embodiments, the least squares method is used as a regression analysis method to generate an approximation line that approximates the feature points DP, but other methods, such as principal component analysis, are used. It's okay to be beaten.

(3) In each of the above embodiments, the road shape recognition device 10 and the road shape determination device 100 are realized by the CPU 101 executing the road shape recognition programs Pr1, Pr2, Pr3. It may also be implemented in hardware by circuits or discrete circuits. That is, the control unit and its method in each of the above embodiments are implemented by a dedicated computer provided by configuring a processor and memory programmed to execute one or more functions embodied by a computer program. May be realized. Alternatively, the controller and techniques described in this disclosure may be implemented by a dedicated computer provided by a processor configured with one or more dedicated hardware logic circuits. Alternatively, the control unit and the method described in the present disclosure are configured by a combination of a processor and memory programmed to perform one or more functions and a processor configured by one or more hardware logic circuits. may be implemented by one or more dedicated computers. The computer program may also be stored as instructions executed by a computer on a computer-readable non-transitory tangible storage medium.

Although the present disclosure has been described above based on the embodiments and modifications, the embodiments of the invention described above are for facilitating understanding of the present disclosure, and do not limit the present disclosure. This disclosure may be modified and improved without departing from its spirit or the scope of the claims, and this disclosure includes equivalents thereof. For example, the embodiments corresponding to the technical features in each form and the technical features in the modifications described in the column of the summary of the invention may be used to solve some or all of the above-mentioned problems, or to solve the above-mentioned problems. In order to achieve some or all of the effects, it is possible to replace or combine them as appropriate. Further, unless the technical feature is described as essential in this specification, it can be deleted as appropriate.

DESCRIPTION OF SYMBOLS 10... Road shape recognition device, 100... Road shape determination device, 101... CPU, 102... Memory, 103... Input/output interface, 104... Bus, 50... Vehicle, Pr1, Pr2, Pr3... Road shape recognition program.

Claims (8)

A road shape recognition device (10),
a feature point acquisition unit (103) that obtains a plurality of feature points indicating the road shape at a predetermined timing;
The recognition distance of the road shape is divided into a plurality of sections by a predetermined division distance, and the section road is divided into each section using the current feature points acquired at the current timing and the past feature points acquired at the past timing. A road shape determination unit (101, Pr1) that generates a shape line and determines a road shape by connecting the generated section road shape lines, wherein the position of the current feature point and the position of the past feature point are determined in advance. a road shape determining unit that generates the section road shape line using an average feature point obtained by averaging the position of the current feature point and the position of the past feature point when the distance is within a predetermined distance range; A road shape recognition device comprising: The road shape recognition device according to claim 1,
The road shape determination unit is a road shape recognition device that corrects the position of the past feature point according to the amount of movement of the vehicle. The road shape recognition device according to claim 1 or 2,
The road shape determination unit determines the reliability of the current feature point and the past feature point during the averaging, and obtains the average feature point by weighted averaging in which feature points with high reliability are weighted. recognition device. The road shape recognition device according to any one of claims 1 to 3,
The road shape determination unit is a road shape recognition device that calculates the average feature point by excluding current feature points that have a large amount of deviation from the section road shape line generated in the past. The road shape recognition device according to claim 4,
The road shape determining unit sets the reliability of the feature points not used in calculating the average feature point to a reliability lower than the reliability of the feature points used in calculating the average feature point. , road shape recognition device. A road shape recognition device (10),
a feature point acquisition unit (103) that obtains a plurality of feature points indicating the road shape at a predetermined timing;
Divide the recognition distance of the road shape into a plurality of sections by the segmentation distance for each of a plurality of predetermined segmentation distances, and generate a segment road shape line for each segment using the acquired feature points. A road shape determination unit (101, Pr2) that determines a road shape by connecting the section road shape lines determined, and selects the segment distance that reduces the error of each of the determined road shapes. A road shape recognition device, comprising: a road shape determination unit that selects a road shape corresponding to the segment distance. A road shape recognition method, comprising:
Acquire multiple feature points indicating the road shape at predetermined timing,
Divide the recognition distance of the road shape into multiple sections using predetermined segmentation distances,
When generating a section road shape line for each section using the current feature point acquired at the current timing and the past feature point acquired at the past timing, the position of the current feature point and the past feature point are If the position is within a predetermined distance range, generating the section road shape line using an average feature point obtained by averaging the position of the current feature point and the position of the past feature point,
A road shape recognition method that determines a road shape by connecting the generated section road shape lines. A road shape recognition method, comprising:
Acquire multiple feature points indicating the road shape at predetermined timing,
Divide the recognition distance of the road shape into a plurality of sections by the segmentation distance for each of a plurality of predetermined segmentation distances , and generate a segment road shape line for each segment using the acquired feature points. When determining a road shape by connecting the section road shape lines, the segment distance that reduces the error of each of the determined road shapes is selected, and the road shape corresponding to the selected segment distance is selected. A road shape recognition method that determines the road shape.

Images