Self trailer boundary dimension detection method based on unmanned container truck perception

Abstract

The invention discloses a trailer boundary dimension detection method based on unmanned container truck perception, and belongs to the technical field of automatic driving and vehicle engineering. In order to solve the problems of low manual correction efficiency and poor flexibility of external measurement equipment in the prior art, the invention provides the following scheme: acquiring three-dimensional point cloud information containing a vehicle through a vehicle-mounted laser radar, calculating a trailer rotation angle in real time by combining a high reflection intensity identifier, dynamically delimiting an interest region by using vehicle-mounted parameters, and filtering out a tractor point cloud; the remaining point clouds are distributed to a two-dimensional grid map, and candidate trailer point clouds are extracted through ground point cloud filtering and grid clustering analysis; finally, the maximum left width, the maximum right width, the maximum length and the maximum height of the trailer are calculated based on segmented time sequence checking, and the accurate boundary size is output. According to the method, manual intervention or external equipment support is not needed, real-time detection can be achieved under any trailer posture, and the safety and efficiency of unmanned container truck operation are effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of automatic driving, and particularly relates to a self-trailer boundary size detection method based on unmanned truck perception. BACKGROUND

[0002] The self-size parameter of an unmanned vehicle is an important reference basis for realizing perception and planning of automatic driving. If the size parameter is smaller than the actual size, path planning may cause collision when passing through a narrow road, and perception may easily misdetect the self. If the size parameter is larger than the actual size, the probability of forced stopping is increased, and the operation efficiency of the unmanned vehicle is affected.

[0003] In the current mainstream unmanned driving solution, the self-size is preset by referring to the vehicle parameter at the factory, but in a closed unmanned truck operation scene, there is often a demand for medium and large goods that exceed the original boundary of the trailer. At this time, the preset size is incorrect, and the trailer boundary size parameter must be corrected. Manual correction mainly relies on the participation of on-site operation and maintenance personnel, uses a tape measure to measure the size value of the goods exceeding the boundary, and then modifies the preset boundary size through a man-machine interaction terminal. Non-contact correction relies on a measurement platform outside the vehicle, uses a combination device of a camera and a displacement sensor to measure the overall length, width and height of the vehicle, and converts the trailer boundary size.

[0004] The manual measurement method has low automation, does not meet the requirements of automatic driving for reducing the number of people, and the operation and maintenance personnel have high work pressure and low work efficiency and are prone to make mistakes when performing high-intensity transportation tasks. The foregoing non-contact solution has high requirements for the posture of the whole vehicle during measurement, the detection location is fixed, and the trailer and the tractor must be on the same axis, so the flexibility is low and real-time detection with the vehicle cannot be achieved. SUMMARY

[0005] The technical problem to be solved by the application is to overcome the deficiencies of the prior art and provide a self-trailer boundary size detection method based on unmanned truck perception. Three-dimensional point cloud information containing the vehicle is obtained by a vehicle-mounted laser radar, the trailer corner is calculated in real time in combination with a high reflection intensity mark, the interest region is dynamically delimited by using the vehicle parameter, and the tractor point cloud is filtered out. The remaining point cloud is distributed to a two-dimensional grid map, candidate trailer point cloud is extracted through ground point cloud filtering and grid clustering analysis, and finally the maximum left width, right width, length and height of the trailer are calculated based on segmented time sequence checking, and accurate boundary size is output.

[0006] The application provides a self-trailer boundary size detection method based on unmanned truck perception, which comprises the following steps: S1. Three-dimensional space point cloud information containing the vehicle is obtained by a vehicle-mounted laser radar, and the point cloud of each laser radar is converted to the vehicle coordinate system of the tractor to obtain original point cloud C0. S2 calculates the angle difference r between the trailer center axis and the towing vehicle center axis in real time by laser radar as the trailer turning angle through high reflection intensity marks installed at the front of the trailer; S3. Based on the trailer turning angle r and the trailer parameters, the range of the region of interest is set, and the original point cloud C0 is screened to obtain a point cloud C1 that does not contain the towing vehicle point cloud; S4. Assign the point cloud C1 to a two-dimensional grid map, construct a two-dimensional grid map M1, and record the point cloud information contained in each grid; S5. The ground point cloud is filtered out by the absolute height difference method and the RANSAC plane fitting method to obtain an updated two-dimensional grid map M2 and the corresponding non-ground point cloud C2; S6. Based on the trailer turning angle r and the trailer parameters, the non-empty grids in M2 are searched and analyzed to obtain a candidate trailer point cloud C3; S7. The trailer point cloud C3 is subjected to segmented timing verification, the maximum left width wl, right width wr, trailer length and height on both sides of the trailer are calculated, and the trailer boundary size information is output.

[0007] As a further technical solution of the present application, in step S2, the calculation of the trailer turning angle specifically includes: S21. Two groups of high reflection intensity marks are installed at the front of the trailer to ensure that the projections of the two groups of marks to the trailer center axis are at the same position; S22. The mark point cloud is selected from the original point cloud C0 according to the reflection intensity; S23. The projection straight line of the mark point cloud in the horizontal plane is fitted by the least square method, and the included angle r1 of the straight line and the positive direction of the towing vehicle X axis is calculated; S24. According to the formula r = r1 - π / 2, the real-time trailer turning angle r is obtained.

[0008] Further, in step S3, the calculation formula of the region of interest range is: Xmin = sx - lb - w / 2 - b1, Xmax = sx + lf + w / 2 + b2, Ymin = sy - lb - w / 2 - b3, Ymax = sy + lf + w / 2 + b4, wherein sx and sy are the projection distances of the trailer hinged point to the towing vehicle coordinate system, lb and lf are the distances from the trailer tail and front to the hinged point, w is the original trailer width, and b1-b4 are range expansion parameters.

[0009] Further, in step S4, the construction of the two-dimensional grid satisfies: Lmap = l × row_num, Wmap = l × col_num, wherein, l is the grid unit size, row_num and col_num are the total number of rows and columns of the grid map respectively, and Lmap>Xmax-Xmin, Wmap>Ymax-Ymin.

[0010] Further, in step S6, the cluster analysis specifically includes: S61. Traversing the non-empty grid in M2, the projection distance dl and the vertical distance dw of each grid to the trailer center axis are calculated; S62. Screening the grid meeting sx-lb≤dl≤sx+lf and -w / 2≤dw≤w / 2 as a candidate trailer range; S63. The grid meeting the conditions is clustered as a candidate trailer point cloud C3 through neighborhood search and threshold judgment.

[0011] Further, in step S7, the segmented timing check includes: S71. Dividing the trailer point cloud C3 into n segments along the center axis direction, and calculating the left width wl, the right width wr and the length upper and lower bounds of each segment; S72. Variance analysis is performed on the calculation results of multiple frames of data, and the abnormal segment with variance exceeding the threshold is discarded, and the iteration is performed until the variance meets the standard; S73. Taking the maximum left width, right width and total length of each segment as the final trailer boundary size.

[0012] Further, the high reflection intensity identifier is a reflective patch or a laser reflection plate, and the installation position ensures that it is within the effective viewing angle range of the laser radar.

[0013] The advantages of the present application are that, The method does not need manual participation and any external road end measuring equipment, uses the laser radar equipped on the autonomous vehicle, combines the trailer corner and the initial size of the vehicle and other vehicle parameters, detects and outputs the trailer boundary size information of the vehicle in real time and high precision, and improves the vehicle passing performance and safety. The trailer point cloud grid clustering method provided by the present application improves the calculation efficiency, effectively retains the original point cloud information of the trailer, can guarantee the detection precision of the trailer boundary size, and meets the real-time performance of the algorithm. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 is the overall flow chart of the trailer boundary size detection method of the present application; Figure 2 is the vehicle coordinate system and the overhead view of the vehicle parameters of the present application; Figure 3A schematic diagram for constructing a two-dimensional grid of the present application; Figure 4 A schematic diagram for a grid clustering process of the present application. DETAILED DESCRIPTION

[0015] Referring to Figure 1 The embodiment provides a self-trailer boundary size detection method based on unmanned truck sensing, and comprises the following steps: S1. Obtain three-dimensional space point cloud information containing the ego vehicle through the vehicle-mounted laser radar, and convert the point cloud of each laser radar to the vehicle coordinate system of the towing vehicle to obtain an original point cloud C0; S2. Use high-reflection-intensity markers installed at the front bumper of the trailer to calculate the angle difference r between the center axis of the trailer and the center axis of the towing vehicle in real time through the laser radar, as the trailer turning angle; S3. Set a region of interest range based on the trailer turning angle r and the vehicle parameters, and screen the original point cloud C0 to obtain a point cloud C1 that does not contain the point cloud of the towing vehicle; S4. Distribute the point cloud C1 to a two-dimensional grid map to construct a two-dimensional grid map M1, and record the point cloud information contained in each grid; S5. Filter out the ground point cloud by using the absolute height difference method and the RANSAC plane fitting method to obtain an updated two-dimensional grid map M2 and the corresponding non-ground point cloud C2; S6. Perform neighborhood search and clustering analysis on the non-empty grids in M2 based on the trailer turning angle r and the vehicle parameters to obtain a candidate trailer point cloud C3; S7. Perform segmented timing verification on the trailer point cloud C3, calculate the maximum left width wl, right width wr, trailer length and height of the trailer on both sides, and output the trailer boundary size information.

[0016] In step S2, the calculation of the trailer turning angle specifically comprises: S21. Install two groups of high-reflection-intensity markers at the front bumper of the trailer to ensure that the projections of the two groups of markers to the center axis of the trailer are at the same position; S22. Screen out the marker point cloud from the original point cloud C0 according to the reflection intensity; S23. Fit the projection straight line of the marker point cloud in the horizontal plane by using the least square method, and calculate the included angle r1 between the straight line and the positive direction of the X-axis of the towing vehicle; S24. Obtain the real-time trailer turning angle r according to the formula r = r1 - π / 2.

[0017] As Figure 2 shown, in step S3, the calculation formula of the region of interest range is: Xmin = sx - lb - w / 2 - b1, Xmax = sx + lf + w / 2 + b2, Ymin = sy - lb - w / 2 - b3, Ymax = sy + lf + w / 2 + b4, Where sx and sy are the projected distances from the trailer articulation point to the tractor coordinate system, lb and lf are the distances from the rear and front of the trailer to the articulation point, respectively, w is the original trailer width, and b1-b4 are range augmentation parameters.

[0018] like Figure 3 As shown, in step S4, the construction of the two-dimensional grid satisfies: Lmap = l × row_num, Wmap = l × col_num, Where l is the raster unit size, row_num and col_num are the total number of rows and columns of the raster map, respectively, and Lmap > Xmax - Xmin, Wmap > Ymax - Ymin.

[0019] In the construction of a two-dimensional grid map, the XOY coordinate system is the vehicle coordinate system of the tractor, and the XgOgYg coordinate system is a Cartesian coordinate system specifically designed for grid maps. Each coordinate corresponds to only one grid cell. For example, g(row,col) means that grid cell g is located in the row-th row and column-th column of the two-dimensional grid map.

[0020] Traverse the C1 point cloud and assign all points to the grid cells of the 2D raster image. For any point p(x,y,z), there is a unique grid cell g(row,col) corresponding to it, that is: row=(x-Xmin) / I⌋ col=(y-Ymin) / I⌋ After the traversal is complete, update all point information to the two-dimensional raster graph M1, and record the number of points contained in each grid of M1, as well as the index (address) of the contained points in C1.

[0021] In step S5, ground point cloud filtering is performed. As the fulcrum supporting the trailer, the ground point cloud is typically included within the region of interest. Considering that the ground is not part of the trailer, the absolute height difference method is used to filter out the ground point cloud to avoid interference. The specific process is as follows: 1. Traverse all grids in the two-dimensional grid map, and based on the radar installation height, select grids whose highest point is below a certain height from the ground plane and whose internal point height difference is within a certain threshold as ground candidates. 2. Based on the Random Sample Consensus (RANSAC) method, planar matching is performed on the point cloud set in the ground candidate grid to obtain the ground point cloud set; 3 Based on the point cloud C1, the ground point cloud set is filtered to obtain the point cloud C2 after filtering the ground, and the grid map M1 is updated to obtain the two-dimensional grid map M2 containing only the non-ground point cloud, and the number and index of the non-ground point cloud contained in each grid in M2.

[0022] In step S6, the grid is clustered. Each non-empty grid in M2 potentially contains a trailer cargo point cloud. Based on the specially designed two-dimensional grid clustering, the relationship between the non-empty grid and the trailer position is determined. The point cloud in the non-empty grid next to the trailer in the physical space is taken as the candidate trailer point cloud C3, and the logical flow is as shown in Figure 4 The specific process is as follows: Traverse the non-empty grid g in the two-dimensional grid map M2. The grid center point coordinates g(x, y) represent its position in the vehicle coordinate system. The projection distance of the grid g to the articulation point is calculated by using the vehicle parameters sx, sy, that is: dx = x - sx dy = y - sy Where dx is the projection distance of the grid center point and the articulation point on the X-axis of the vehicle coordinate system, and dy is the projection distance of the grid center point and the articulation point on the Y-axis of the vehicle coordinate system.

[0023] The real-time trailer turning angle r calculated in step two is used to calculate the projection distance dl of the grid g along the trailer center axis to the articulation point, and the vertical distance dw of the grid g to the trailer center axis, that is: dl = dx * cos(r) + dy * sin(r) dw = dy * cos(r) - dx * sin(r) The vehicle parameters lb, lf, w are used to determine whether the projection distance satisfies the following conditions: sx - lb <= dl <= sx + lf -w / 2 <= dw <= w / 2 The grid that satisfies the condition is considered to be within the candidate trailer range.

[0024] For the grid g within the candidate trailer range, the neighborhood search is performed along the positive and negative directions of the Xg and Yg of the grid coordinate system Xg and Yg by a certain number of steps p. The total number of points contained in the grid g and its neighborhood grid set is calculated. The grid g whose total number of points satisfies the threshold value t is added to the initial seed grid set (the number of steps p and the threshold value t need to be adjusted according to the performance indicators of the real vehicle laser radar).

[0025] Traverse the neighborhood grid set of the seed grid to re-perform neighborhood search with the neighborhood grid as the center. The neighborhood grid whose total number of points satisfies the threshold value t is added to the seed grid set; Mark the grid that has been judged whether it belongs to the seed grid set to prevent repeated traversal; Repeat step 3, 4 until the number of grids in the seed grid set no longer increases; According to the index of each point in the seed grid in C2, restore to point cloud C3.

[0026] Point cloud C3 is the complete trailer cluster point cloud at the current time.

[0027] In step S7, trailer boundary calculation and size output. Trailer boundary size detection requirements often appear in trailer loading and unloading stage, at this time, the complex working environment, personnel movement and machine work caused dust and other interference factors, may affect the trailer point cloud clustering results obtained in step six. The segmented size timing check of trailer cluster point cloud is carried out to reduce external interference, the specific process is as follows: 1. The trailer cluster point cloud is divided into n segments along the trailer center axis direction, wherein the length of the second segment to the n-1 segment is consistent, and the first segment and the n segment contain all the trailer point clouds in the front and rear directions; Combined with the trailer corner r, the left width wl and the right width wr of the trailer point cloud are calculated, that is: wl=|max((yi–sy)*cos(r)-(xi–sx)*sin(r))| wr=|min((yi–sy)*cos(r)-(xi–sx)*sin(r))| Where xi, yi is the coordinates of the i-th point in the segment trailer point cloud in the vehicle coordinate system of the towing vehicle.

[0028] For the first segment and the n segment of the trailer point cloud, the upper limit l_max and the lower limit l_min of the point cloud are additionally calculated, that is: l_max=max((xi–sx)*cos(r)+(yi–sy)*sin(r)) l_min=min((xi–sx)*cos(r)+(yi–sy)*sin(r)) Repeat the above process to record the left width and right width values of each segment of the trailer point cloud in a period of time, and calculate the variance; For the trailer point cloud segment with variance greater than the threshold value, discard the current result and repeat steps 2 and 3 until the variance is less than the threshold value; Taking the maximum left width value and the maximum right width value in each segment of the trailer point cloud, the maximum vertical distance from the trailer center axis to the trailer side can be obtained. Taking the difference between the upper limit l_max of the first segment of the trailer point cloud and the lower limit l_min of the n segment of the trailer point cloud, the length of the trailer can be obtained. Taking the highest point height in each segment of the trailer point cloud, the real-time height of the trailer in the vehicle coordinate system of the towing vehicle can be obtained.

[0029] High reflection intensity is identified as a reflective patch or a laser reflection plate, and its installation position ensures that it is within the effective viewing angle range of the laser radar.

[0030] The foregoing merely illustrates the principles of the application and application of its more prominent features. Those skilled in the art will appreciate that the application is not limited to the specific embodiments described herein, which are presented for the purpose of illustration and not limitation. The present application is limited only by the claims and their equivalents.

Claims

1. A method for detecting the boundary dimensions of a trailer based on the perception of an unmanned truck, characterized in that, Includes the following steps: S1. Obtain three-dimensional spatial point cloud information containing the vehicle through the vehicle-mounted lidar, and transform the point clouds of each lidar into the vehicle coordinate system of the tractor to obtain the original point cloud C0. S2 uses a high-reflectivity marker installed on the front windshield of the trailer and calculates the angle difference r between the centerline of the trailer and the centerline of the tractor in real time using lidar, which is used as the trailer turning angle. S3. Based on the trailer turning angle r and the accompanying parameters, set the range of the region of interest, and filter the original point cloud C0 to obtain a point cloud C1 that does not contain the tractor point cloud; S4. Assign point cloud C1 to a two-dimensional raster map, construct a two-dimensional raster map M1, and record the point cloud information contained in each raster; S5. The ground point cloud is filtered out by the absolute height difference method and the RANSAC plane fitting method to obtain the updated two-dimensional raster map M2 and the corresponding non-ground point cloud C2. S6. Based on the trailer turning angle r and accompanying parameters, perform neighborhood search and cluster analysis on the non-empty grids in M2 to obtain candidate trailer point clouds C3; S7. Perform segmented timing verification on trailer point cloud C3, calculate the maximum left width wl, right width wr, trailer length and height on both sides of the trailer, and output trailer boundary dimension information.

2. The method for detecting the boundary dimensions of a trailer based on unmanned truck perception according to claim 1, characterized in that, In step S2, the calculation of the trailer turning angle specifically includes: S21. Install two sets of high-reflectivity markers on the front windshield of the trailer, ensuring that the projections of the two sets of markers onto the centerline of the trailer are at the same position; S22. Select the identifier point cloud from the original point cloud C0 based on the reflection intensity; S23. Fit the projection line of the marker point cloud onto the horizontal plane using the least squares method, and calculate the angle r1 between the line and the positive direction of the X-axis of the tractor. S24. According to the formula r = r1 - π / 2, the real-time trailer turning angle r is obtained.

3. The method for detecting the boundary dimensions of a trailer based on unmanned truck perception according to claim 1, characterized in that, In step S3, the formula for calculating the region of interest is: Xmin = sx - lb - w / 2 - b1, Xmax = sx + lf + w / 2 + b2, Ymin = sy - lb - w / 2 - b3, Ymax = sy + lf + w / 2 + b4, Where sx and sy are the projected distances from the trailer articulation point to the tractor coordinate system, lb and lf are the distances from the rear and front of the trailer to the articulation point, respectively, w is the original trailer width, and b1-b4 are range augmentation parameters.

4. The method for detecting the boundary dimensions of a trailer based on unmanned truck perception according to claim 1, characterized in that, In step S4, the construction of the two-dimensional grid satisfies: Lmap = l × row_num, Wmap = l × col_num, Where l is the raster unit size, row_num and col_num are the total number of rows and columns of the raster map, respectively, and Lmap > Xmax - Xmin, Wmap > Ymax - Ymin.

5. The method for detecting the boundary dimensions of a trailer based on unmanned truck perception according to claim 1, characterized in that, In step S6, the cluster analysis specifically includes: S61. Traverse the non-empty grids in M2 and calculate the projected distance dl and vertical distance dw from each grid to the trailer centerline; S62. Select the grid cells that satisfy sx - lb ≤ dl ≤ sx + lf and -w / 2 ≤ dw ≤ w / 2 as candidate trailer ranges; S63. Through neighborhood search and threshold judgment, the raster that meets the conditions is clustered into candidate trailer point cloud C3.

6. The method for detecting the boundary dimensions of a trailer based on unmanned truck perception according to claim 1, characterized in that, In step S7, the segmented timing verification includes: S71. Divide the trailer point cloud C3 into n segments along the central axis, and calculate the left width wl, right width wr, and upper and lower bounds of the length of each segment; S72. Perform variance analysis on the calculation results of multi-frame data, discard outlier segments with variance exceeding the threshold, and iterate until the variance reaches the target. S73. Take the maximum left width, right width and total length of each segment as the final trailer boundary dimensions.

7. The method for detecting the boundary dimensions of a trailer based on unmanned truck perception according to claim 1, characterized in that, The high reflectivity indicator is a reflective patch or laser reflector, and its installation position is ensured to be within the effective field of view of the lidar.

Images