Dynamic vehicle real-time detection method and device based on roadside laser radar

Abstract

The invention discloses a dynamic vehicle real-time detection method and device based on a roadside laser radar, and the method specifically comprises the steps: carrying out the stationary point filtering and the outlier filtering of a self-adaptive threshold in a matrix with a spherical coordinate attribute through combining point cloud data of different time and space, and carrying out the clustering segmentation of target point cloud data in a space rectangular coordinate system, and a vehicle target is identified according to the point cloud number features and the range features. After dimension reduction is carried out on the plane, angular points and an initial main direction are determined according to the characteristics of the L shape and the I shape of the vehicle, a mathematical model of a bounding box is constructed and designed through the area criterion and the distance criterion of contour points in combination with the initial main direction, and the optimal bounding box is searched through iteration. And finally, creating a line buffer area of the planar bounding box, and verifying whether mixed noisy points exist. According to the method, noise interference is effectively reduced, the influence of appearance change on vehicle recognition is relieved, the algorithm is more flexible and universal by calculating the adaptive filtering threshold value and the optimal solution model, and the accuracy of real-time vehicle detection is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation, and in particular to a method and device for real-time detection of dynamic vehicles based on roadside laser radar. Background Art

[0002] In intelligent traffic management, real-time detection of road vehicles has important application value. It can help us count traffic flow, ensure road construction safety, monitor traffic violations, and support autonomous vehicle testing. At the same time, complex road environments and massive data have put forward higher requirements for technical solutions, requiring operational flexibility, high detection accuracy, real-time dynamic detection capabilities, and high efficiency of data transmission. In existing technologies, LiDAR has been widely used in the field of intelligent transportation due to its ability to obtain high-precision three-dimensional spatial information, excellent depth perception, and strong anti-interference characteristics.

[0003] LiDAR-based target detection methods include OBB bounding box method, convex hull fitting, deep learning, etc. However, these methods still have some challenges, such as high computational complexity, serious information loss, and noise sensitivity. At present, in the scenario of real-time roadside detection of dynamic vehicles, there is still a lack of a technical solution that can efficiently obtain target geometric features, provide timely feedback on abnormal information, and is suitable for real-time detection. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art. In view of the shortcomings of the prior art, a method and device for real-time detection of dynamic vehicles based on roadside laser radar are provided.

[0005] In order to achieve the above objectives, the technical solution of the present invention is as follows: In the first aspect, a dynamic vehicle real-time detection method based on roadside laser radar comprises the following steps:

[0006] Step 1: Obtain 3D point cloud data of real-time road conditions;

[0007] Step 2: Perform stationary point filtering and outlier point filtering on the three-dimensional point cloud data;

[0008] Step 3: cluster the filtered 3D point cloud data to obtain single target point cloud data, and determine whether it is a dynamic vehicle, and finally obtain vehicle point cloud data;

[0009] Step 4, calculating the corner points of the vehicle point cloud data, segmenting the cluster of the vehicle front part according to the corner points, and then fitting to obtain the initial width main direction of the vehicle;

[0010] Step 5: Calculate the boundary contour points of the vehicle, combine the initial width main direction of the vehicle, take the minimum sum of the area weight and distance weight of the bounding box as the objective function, build the mathematical model of the bounding box, take the initial width main direction slope of the vehicle as the design variable, iteratively find the optimal bounding box, and filter the mixed noise points;

[0011] Step 6: Calculate the geometric features of the vehicle based on the optimal bounding box obtained in step 5 and output the detection results.

[0012] Furthermore, in step one, three-dimensional point cloud data of the real-time road conditions are obtained by laser radar, and the laser radar system equipment is arranged outside the motor vehicle road at a distance of L m, wherein the direction of the laser radar is opposite to the direction of vehicle travel, and the center line of the laser radar is parallel to the lane line or toward the lane.

[0013] Furthermore, in step 2, the static point filtering method is to subtract the currently acquired point cloud data S from the point cloud data S1 separated by a fixed frame. 11 The outlier filtering adopts radius filtering based on continuous search, and the threshold of radius filtering has adaptive characteristics, and its calculation formula is:

[0014] T i,j =R i,j ·f·E;

[0015] Where, T i,j ,R i,j is the radius threshold and radial distance of the point in the i-th row and j-th column in the matrix, f is the minimum resolution of the distance analysis formula; E is the correction coefficient for the change of point cloud sparsity, and the relative volume fraction indicator is used for experimental testing based on the radial distance of the vehicle, and the correction coefficient is set according to the expected value.

[0016] Furthermore, in step three, the method for determining a dynamic vehicle is to reduce the dimension of the clustered target point cloud data to the two-dimensional plane to which the ground belongs, and then calculate the quantity characteristics and range characteristics of the point cloud. If the following formula is satisfied, the target point cloud data is dynamic vehicle data;

[0017]

[0018] Where Q is the number of point clouds, r is the horizontal viewing angle resolution of the lidar, and E is the correction coefficient for the sparsity change of the point cloud; MaxY is the maximum value of the Y coordinate, MinY is the minimum value of the Y coordinate, the Y coordinate value reflects the lane width, and MaxY MinY reflects the vehicle width range.

[0019] Furthermore, in step 4, the corner point is determined by the minimum radial distance of the point cloud; then, a straight line L is drawn through the origin and the corner point, and the straight line L divides the vehicle point cloud data into class A cluster and class B cluster. Class B cluster is the cluster of the front part of the vehicle. The straight line L2 of class B cluster after RANSAC fitting represents the initial width main direction of the vehicle, and class A cluster is the point cloud of the side of the vehicle.

[0020] Further, in step 5, the mathematical model of the optimal bounding box design refers to taking the slope of the straight line L2 as the design variable, taking the sum of the area weight and the distance weight of the bounding box as the objective function, and finding the value of the optimal function within a suitable range, which is specifically expressed as follows:

[0021] Min W=w1Area+w2Distance

[0022] Area = a·b

[0023]

[0024]

[0025] Where W is the objective function; Area is the area of ​​the bounding box, which is normalized; a is the width of the bounding box, b is the length of the bounding box; Distance is the sum of the shortest distances from all contour points to the four straight lines of the bounding box, which is normalized; J is the number of contour points; d i is the minimum value of the distance from contour point i to the four straight lines; w1 and w2 are weights respectively; k2 is the initial slope of L2, and k is the search step size.

[0026] Furthermore, in step five, the mixed noise is filtered by creating a line buffer, gradually approaching the four straight lines constituting the optimal bounding box to the inside of the bounding box, and determining whether mixed noise is contained in the continuously changing line buffer.

[0027] Furthermore, in step six, the geometric features refer to the length, width, height, and skew angle arctan(k2) of the vehicle, and the output result refers to uploading the geometric features of the vehicle to the cloud platform and storing and displaying them.

[0028] In the second aspect, the present invention also provides a dynamic vehicle real-time detection device based on roadside laser radar, including a laser radar, a memory, a processor, a PC, and a network interface. The memory stores executable code, and is characterized in that when the processor executes the executable code, the dynamic vehicle real-time detection method based on laser radar is implemented.

[0029] In a third aspect, the invention also provides a computer-readable storage medium having a program stored thereon, characterized in that when the program is executed by a processor, the method for real-time detection of dynamic vehicles based on roadside lidar is implemented.

[0030] Beneficial effects of the present invention:

[0031] The present invention can effectively reduce noise interference through stationary point filtering, outlier point filtering, and a mixed point filtering method based on buffer path search. By taking advantage of the fact that vehicle point clouds are prone to present L-shape and I-shape, the front part of the vehicle is used as the initial main direction for fitting, which effectively alleviates the impact of appearance changes on vehicle recognition and improves the efficiency and robustness of the algorithm. By calculating the adaptive filtering threshold and the optimal solution model, the algorithm is made more flexible and universal, and the accuracy of real-time vehicle detection is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 The figure is a schematic block diagram of the process structure of an embodiment of the present invention.

[0034] Figure 2 This is a schematic diagram of the layout of the laser radar detection system according to an embodiment of the present invention.

[0035] Figure 3 Schematic diagram of the initial main direction of a vehicle according to an embodiment of the present invention.

[0036] Figure 4 Schematic diagram of a line buffer according to an embodiment of the present invention

[0037] Figure 5 This is a diagram showing the vehicle detection effect of an embodiment of the present invention.

[0038] Figure 6 This is a structural diagram of a dynamic vehicle real-time detection device based on laser radar in the present invention. DETAILED DESCRIPTION

[0039] In order to more clearly illustrate the technical problems, technical solutions and beneficial effects to be solved by the present application, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0040] The following is a further detailed description through specific implementation methods:

[0041] Example

[0042] This embodiment provides a method and device for real-time detection of dynamic vehicles based on roadside laser radar. Figure 1 As shown, the following steps are included:

[0043] Step 1: Obtain 3D point cloud data of the real-time road conditions. The laser radar system equipment is arranged outside the motor vehicle road, 1m away, with the laser radar facing in the opposite direction of the vehicle, the center line of the laser radar parallel to the lane line, and the origin of the coordinate system is the laser radar position, such as Figure 2 As shown; when acquiring data, in addition to recording the point cloud data at the current moment, the data of the previous 10 frames are retained, and the total data can be expressed as: P = [S1, S2, S3......S 11 ], where S 11 For the current frame, each frame of point cloud data is stored in a matrix with spherical coordinate attributes. The dimension of the matrix is ​​300×125. The physical meaning of the number of rows 300 represents the horizontal viewing angle range, and the physical meaning of the number of columns 125 represents the vertical viewing angle range. The matrix stores the radial distance of the point cloud.

[0044] Step 2: The currently acquired point cloud data S 11 Perform stationary point filtering and outlier point filtering, and then convert from the spherical coordinate system to the spatial rectangular coordinate system. Stationary points include ground points, green belts, fences, etc. Subtract the currently acquired point cloud data S from the point cloud data S1 11 , if the difference is less than 0.3m, the value is assigned to 0; the outlier filtering adopts a radius filtering method based on continuous search, for example, in the same level (column) of the matrix S[i,1], (i=1,2,3,......m), the radius threshold of point p1 is defined as T1, if the Euclidean distance from the next point p2 to this point is greater than the radius threshold, then p2 is an interval point, and then with p2 as the center, the radius threshold is T2, and the search and judgment are continued. If 3 interval points appear continuously, p1, p2, and p3 are filtered as noise points. The threshold is calculated as follows:

[0045] T i,j =R i,j ·f·E;

[0046] Where, T i,j ,R i,jare the radius threshold and radial vector distance of the point in the i-th row and j-th column in the matrix, f is the minimum resolution of the lidar distance analytical formula of 0.01m, E is the correction coefficient for the sparsity change of the point cloud, and when the radial distance interval is [5,10], [10,20], [20,30], [30,40], [40,50], [50,60], [60,70], [70,80], the values ​​of E are 1, 2, 3, 4, 5, 5, 5, respectively. When it is within 5m, it is not detected.

[0047] The correction coefficient E is determined by using the relative volume fraction indicator for experimental testing. 200 frames of point cloud data are taken in each radial distance segment, and different E are introduced to calculate the relative volume fractions of the point cloud target after clustering segmentation and the actual target. The best expected value of the test data result is taken. The larger the recent expectation, the more suitable E is.

[0048] Step three, in the spatial rectangular coordinate system, use the DBSCAN density clustering algorithm to cluster and segment the three-dimensional point cloud data, obtain single target point cloud data, and determine whether it is a dynamic vehicle. When clustering, set the DBSCAN field radius to 1, the minimum number of points within the field radius to 5, and obtain single target point cloud data. The number of target point clouds is 1168, and it is reduced to the two-dimensional plane belonging to the ground, and the minimum and maximum values ​​of the Y coordinates are calculated. If the following requirements are met, the target point cloud data is dynamic vehicle data;

[0049]

[0050] Where Q is the number of point clouds, r is the horizontal viewing angle resolution of the lidar 0.2°, and E is the correction coefficient for the sparsity change of the point cloud; MaxY is the maximum value of the Y coordinate 5.82, MinY is the minimum value of the Y coordinate 3.98, and the Y coordinate value reflects the lane width distance.

[0051] Step 4: Calculate the corner points in the vehicle point cloud data, segment the clusters of the front part according to the corner points, and use the RANSAC random sampling consistency method to fit the initial main direction of the vehicle. The corner points are determined by the minimum distance R value of the points, and the coordinates are (4.59, 10.32); draw a straight line L through the origin and the corner points. The straight line L divides the vehicle point cloud data into clusters A and B. Cluster B is the cluster of the front part of the vehicle. The straight line L2 after RANSAC fitting of cluster B represents the initial width main direction of the vehicle, as shown in Figure 3 As shown, L2 can be expressed as: y2=0.23x+9.07.

[0052] Step 5: Calculate the boundary contour points of the vehicle based on the Alpha Shapes method, combine the initial main direction of the vehicle, build a mathematical model for designing the bounding box, and iteratively find the optimal bounding box. The mathematical model for the optimal bounding box design refers to using the slope k of the straight line L2 as the design variable, and the sum of the area weight and distance weight of the bounding box as the objective function, and finding the value of the optimal function within a suitable range, which can be expressed as:

[0053] Min W=w1Area+w2Distance

[0054] Area = a·b

[0055]

[0056]

[0057] Where W is the objective function; Area is the area of ​​the bounding box, which is normalized; a is the width of the bounding box, which is obtained by calculating the Euclidean distance between the projection point of the contour point on the line L2 and the projection boundary point; b is the length of the bounding box, which is obtained by calculating the Euclidean distance between the projection point of the contour point on the line L1 and the projection boundary point, and the line L1 is perpendicular to the line L2; Distance is the sum of the shortest distances from all contour points to the four lines of the bounding box, which is normalized; J is the number of contour points; d i is the minimum value of the distance from a contour point to the four straight lines; w1 and w2 are 0.4 and 0.6 respectively; the search step of k is 0.02; k2 is the initial slope of L2 0.23. After calculation, the optimal value of k2 is 0.0429.

[0058] Then, a line buffer of the optimal bounding box is created to filter the mixed noise points close to the vehicle. The line buffer is a geometric area formed by offsetting the four straight lines that make up the optimal bounding box into the bounding box. Taking L3 as an example, the line buffer of L3 is as follows: Figure 4 As shown, the filtering steps are as follows:

[0059] In the first step, L3 is offset inward by 0.05m to form a buffer zone f1 with the original bounding box. The number of points falling in the buffer zone f1 is recorded, including those on the line.

[0060] In the second step, repeat the above step (1), offsetting 0.05m each time, and forming buffers f2, f3, etc. with the original bounding box. If the number of points falling in the buffer is greater than 10-D / 10, stop buffering. D is the minimum value of the point cloud X coordinate, 10.26. The X coordinate value reflects the lane length. Therefore, stop buffering when the number of points in the buffer exceeds 9.

[0061] In the third step, if the number of buffering times is greater than 1, the mixed noise points in the buffer are filtered. If the number of other points within the 0.1m radius of the point in the buffer is less than 3, and no new points fall into the buffer when the buffer is expanded next time, the point is judged to be a mixed noise point. After the mixed noise point is identified, it is filtered out.

[0062] Step 6: Calculate the geometric features of the vehicle based on the optimal bounding box obtained in step 5 and output the result. The geometric features refer to the length, width, height, and skew angle of the vehicle. After calculation, the length is 4.3m, the width is 1.90m, the height is 1.58m, and the skew angle is 2.5°, where the skew angle is defined as arctan(k2).

[0063] Each point cloud target after cluster segmentation is implemented according to steps 4 to 6. At the same time, the effect of processing multiple vehicle targets is as follows: Figure 5 shown.

[0064] In order to verify the superiority of the method proposed in the present invention, experiments based on the OBB bounding box model and the traditional L model were conducted on continuous data sets 1-10, and compared with the method proposed in the present invention. The data sets were collected from actual road conditions, and each data set contained 150 frames of data of the same vehicle from far to near.

[0065] Table 1 shows the width errors under three vehicle feature extraction algorithms. The width error reflects the degree of deviation between the calculated width and the actual vehicle width, and the error is expressed as follows:

[0066] δ=W th(A) -W th(P)

[0067] Where δ is the error, W th(A) is the actual width of the vehicle, W th(P) To calculate the width.

[0068] Table 1

[0069]

[0070] The present invention also provides an embodiment of a dynamic vehicle real-time detection device based on a roadside laser radar, and the detection device corresponds to the above-mentioned dynamic vehicle real-time detection method based on a laser radar.

[0071] See Figure 6 The present invention provides a dynamic vehicle real-time detection device based on roadside laser radar, including a laser radar, a memory, a processor, a PC, and a network interface. The memory stores the collected data and stores executable codes. The processor is used to implement a dynamic vehicle real-time detection method based on laser radar in the above embodiment. The PC refers to any computer with data processing capabilities and display.

[0072] An embodiment of a dynamic vehicle real-time detection device based on a roadside laser radar provided by the present invention is implemented by a combination of software and hardware. The operating instructions can be transmitted to the processor by the PC, or the processor can directly execute the internal processor instructions. The implementation process of each unit and module is specifically detailed in the corresponding steps in the above-mentioned dynamic vehicle real-time detection method based on a roadside laser radar, which will not be repeated here.

[0073] The device embodiments described above can determine the spatial position of each unit according to actual needs. In addition to the laser radar, memory, processor, PC terminal, network interface and other hardware, any device with data processing capability in the embodiment where the device is located can also include other hardware according to the actual function of the device with data processing capability, which will not be described in detail. Those skilled in the art can understand and implement it without creative work.

[0074] The present invention also provides a computer-readable storage medium, which can be either non-volatile or volatile. The computer-readable storage medium stores instructions for a real-time detection method for dynamic vehicles based on a roadside laser radar, which can be executed by a computer. The specific working process of the system can refer to the corresponding steps in the aforementioned method embodiment. The computer-readable storage medium can be used as an internal storage unit of the PC in the above-mentioned device embodiment, or as any external storage device with data processing capabilities, which can include a mobile hard disk, an SD card, an optical disk, etc.

[0075] The above description only represents a preferred embodiment of the present invention, and does not limit the present invention. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A real-time dynamic vehicle detection method based on roadside laser radar, characterized in that: The method comprises the following steps: Step 1: Obtain 3D point cloud data of real-time road conditions; Step 2: Perform stationary point filtering and outlier point filtering on the three-dimensional point cloud data; Step 3: cluster the filtered 3D point cloud data to obtain single target point cloud data, and determine whether it is a dynamic vehicle, and finally obtain vehicle point cloud data; Step 4, calculating the corner points of the vehicle point cloud data, segmenting the cluster of the vehicle front part according to the corner points, and then fitting to obtain the initial width main direction of the vehicle; Step 5: Calculate the boundary contour points of the vehicle, combine the initial width main direction of the vehicle, take the minimum sum of the area weight and distance weight of the bounding box as the objective function, build the mathematical model of the bounding box, take the initial width main direction slope of the vehicle as the design variable, iteratively find the optimal bounding box, and filter the mixed noise points; Step 6: Calculate the geometric features of the vehicle based on the optimal bounding box obtained in step 5 and output the detection results.

2. A method for real-time dynamic vehicle detection based on roadside laser radar according to claim 1, characterized in that: In step one, three-dimensional point cloud data of the real-time road conditions is obtained by laser radar. The laser radar system equipment is arranged outside the motor vehicle road at a distance of L m, wherein the direction of the laser radar is opposite to the direction of vehicle travel, and the center line of the laser radar is parallel to the lane line or toward the lane.

3. The method for real-time dynamic vehicle detection based on roadside laser radar according to claim 1, characterized in that: In step 2, the static point filtering method is to subtract the currently acquired point cloud data S from the point cloud data S1 separated by a fixed frame. 11 The outlier filtering adopts radius filtering based on continuous search, and the threshold of radius filtering has adaptive characteristics, and its calculation formula is: T i,j =R i,j ·f·E; Where, T i,j ,R i,j is the radius threshold and radial distance of the point in the i-th row and j-th column in the matrix, f is the minimum resolution of the distance analysis formula; E is the correction coefficient for the change of point cloud sparsity, and the relative volume fraction indicator is used for experimental testing based on the radial distance of the vehicle, and the correction coefficient is set according to the expected value.

4. The method for real-time dynamic vehicle detection based on roadside laser radar according to claim 3 is characterized in that: In step three, the method for determining a dynamic vehicle is to reduce the dimension of the clustered target point cloud data to the two-dimensional plane of the ground, and then calculate the quantity characteristics and range characteristics of the point cloud. If the following formula is satisfied, the target point cloud data is dynamic vehicle data; MaxY-MinY>1.2; Where Q is the number of point clouds, r is the horizontal viewing angle resolution of the lidar, and E is the correction coefficient for the sparsity change of the point cloud; MaxY is the maximum value of the Y coordinate, MinY is the minimum value of the Y coordinate, the Y coordinate value reflects the lane width, and MaxY MinY reflects the vehicle width range.

5. The method for real-time dynamic vehicle detection based on roadside laser radar according to claim 1, characterized in that: In step 4, the corner point is determined by the minimum radial distance of the point cloud; then, a straight line L is drawn through the origin and the corner point, and the straight line L divides the vehicle point cloud data into class A cluster and class B cluster. Class B cluster is the cluster of the front part of the vehicle. The straight line L2 of class B cluster after RANSAC fitting represents the initial width main direction of the vehicle, and class A cluster is the point cloud of the side of the vehicle.

6. A method for real-time dynamic vehicle detection based on roadside laser radar according to claim 5, characterized in that: In step 5, the mathematical model of the optimal bounding box design refers to taking the slope of the straight line L2 as the design variable, taking the sum of the area weight and the distance weight of the bounding box as the objective function, and finding the value of the optimal function within a suitable range, which is specifically expressed as follows: Min W=w1Area+w2Distance Area = a·b Where W is the objective function; Area is the area of ​​the bounding box, which is normalized; a is the width of the bounding box, b is the length of the bounding box; Distance is the sum of the shortest distances from all contour points to the four straight lines of the bounding box, which is normalized; J is the number of contour points; d i is the minimum value of the distance from contour point i to the four straight lines; w1 and w2 are weights respectively; k2 is the initial slope of L2, and k is the search step size.

7. The method for real-time dynamic vehicle detection based on roadside laser radar according to claim 1, characterized in that: In step five, the mixed noise is filtered by creating a line buffer, gradually moving the four straight lines constituting the optimal bounding box closer to the inside of the bounding box, and determining whether there are mixed noise points in the continuously changing line buffer.

8. The method for real-time dynamic vehicle detection based on roadside laser radar according to claim 1, characterized in that: In step six, the geometric features refer to the length, width, height, and skew angle arctan(k2) of the vehicle, and the output result refers to uploading the geometric features of the vehicle to the cloud platform and storing and displaying them.

9. A dynamic vehicle real-time detection device based on roadside laser radar, comprising a laser radar, a memory, a processor, a PC, and a network interface, wherein the memory stores executable code, characterized in that: When the processor executes the executable code, a dynamic vehicle real-time detection method based on a roadside laser radar as described in any one of claims 1-8 is implemented.

10. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by the processor, a real-time dynamic vehicle detection method based on a roadside laser radar as described in any one of claims 1 to 8 is implemented.

Images