Vehicle hitching yaw angle determination method and device, computer device, and storage medium

Abstract

The application relates to a vehicle hitch drift angle determination method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring first point cloud data, the first point cloud data being point cloud data of a current frame of a current vehicle; acquiring historical straight line parameters and determining a search range according to the historical straight line parameters, the historical straight line parameters being parameters of an edge line of a hitch of a historical frame of the current vehicle and determined according to point cloud data of the historical frame of the current vehicle; determining an optimal straight line pair according to the search range, the width of the hitch and the first point cloud data, and determining second point cloud data according to the optimal straight line pair; determining a target straight line according to the second point cloud data, and determining a drift angle according to the target straight line. The application utilizes the historical straight line parameters to determine the search range of the straight line, thereby reducing the search range and accelerating the search speed. In addition, the application searches for a pair of edge lines on both sides of the hitch in combination with the width of the hitch, so that the problem of interference caused by other noise points can be effectively reduced.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, computer equipment, and storage medium for determining the yaw angle of a vehicle trailer. Background Technology

[0002] In many scenarios, certain autonomous vehicles (vehicles consisting of a tractor and a trailer, such as logistics trucks) need to determine the trailer's yaw angle in real time to help complete precise driving or movement.

[0003] In traditional technology, the point cloud data of vehicles is collected in real time, filtered, converted into a set of points on a two-dimensional plane, and then fitted into a straight line to determine the included angle.

[0004] However, in traditional solutions, point cloud filtering is based solely on whether the point coordinates fall within the distance from the origin of the coordinate system to the edge of the trailer—that is, filtering is based solely on the vehicle's size information. This approach is susceptible to interference from side point clouds formed by other vehicles in the vicinity. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, computer equipment, and storage medium for determining the yaw angle of a vehicle trailer in response to the above-mentioned technical problems.

[0006] A method for determining the yaw angle of a vehicle trailer, the method comprising:

[0007] Acquire the first point cloud data, which is the point cloud data of the current vehicle in the current frame;

[0008] Obtain historical straight line parameters and determine the search range based on the historical straight line parameters. The historical straight line parameters are the parameters of the edge line of the vehicle trailer in the current historical frame, which are determined based on the point cloud data of the current vehicle historical frame.

[0009] Based on the search range, the width of the vehicle trailer, and the first point cloud data, determine the optimal straight line pair, and based on the optimal straight line pair, determine the second point cloud data;

[0010] The target straight line is determined based on the second point cloud data, and the yaw angle is determined based on the target straight line.

[0011] In one embodiment, the aforementioned historical frame includes the previous frame, the aforementioned historical straight line parameters include parameters of the target edge lines among the side edge lines of the trailer in the previous frame, and the aforementioned determination of the search range based on the historical straight line parameters includes:

[0012] Based on the parameters of the target edge line of the vehicle in the previous frame and the preset threshold, the range of parameters of the target edge line in the current frame is determined, and the search range is obtained.

[0013] In one embodiment, the process of determining the optimal line pair based on the search range, the width of the vehicle trailer, and the first point cloud data, and then determining the second point cloud data based on the optimal line pair, includes:

[0014] Project the 3D point cloud corresponding to the first point cloud data onto a 2D plane to obtain the projected point cloud;

[0015] Based on the search range and the preset search interval, determine the parameters of each predicted line;

[0016] Based on the predicted straight line parameters and the current vehicle's trailer width, multiple straight line pairs are determined.

[0017] The mass of each straight line pair is determined based on the points in the projected point cloud that are within a first preset range from each straight line pair.

[0018] Determine the optimal line pair based on the quality of each line pair;

[0019] Based on the projected point cloud, the point within the second preset range of the optimal straight line pair is determined, and the second point cloud data is obtained.

[0020] In one embodiment, the above-mentioned process of projecting the three-dimensional point cloud corresponding to the first point cloud data onto a two-dimensional plane to obtain the projected point cloud includes:

[0021] Based on the first point cloud data and the pre-trained graph neural network model, the third point cloud data is obtained. The graph neural network model is used to perform non-linear transformation on the first point cloud data to obtain the category label value of each point, so as to filter each point according to the category label value of each point to obtain the third point cloud data.

[0022] The 3D point cloud corresponding to the third point cloud data is projected onto a 2D plane to obtain the projected point cloud.

[0023] In one embodiment, the above method further includes:

[0024] Based on the coordinates of each point in the first point cloud data, the current vehicle pose, and the current vehicle size information, the first point cloud data is filtered to obtain the fourth point cloud data.

[0025] The third point cloud data, obtained from the first point cloud data and the pre-trained graph neural network model, includes:

[0026] The third point cloud data is obtained based on the fourth point cloud data and the pre-trained graph neural network model.

[0027] In one embodiment, determining the quality of each line pair based on points in the projected point cloud that are within a first preset range from each line pair includes:

[0028] Project the projection point cloud of each line pair within the first preset range onto the corresponding line pair to obtain each target projection point on each line pair.

[0029] The quality of each straight line pair is determined based on the continuity relationship formed by the target projection points on each straight line pair.

[0030] In one embodiment, the above-mentioned determination of the target straight line based on the second point cloud data and the determination of the yaw angle based on the target straight line include:

[0031] The second point cloud data is sampled a preset number of times according to the random sampling consensus algorithm.

[0032] The point cloud data extracted in each iteration is fitted with a straight line using the least squares method to determine the straight line corresponding to each iteration.

[0033] The quality of the line corresponding to each iteration is determined based on the point cloud data extracted in each iteration, and the target line is determined based on the quality of the line corresponding to each iteration.

[0034] The yaw angle is obtained by the angle between the target straight line and the centerline of the current vehicle's front end.

[0035] A device for determining the yaw angle of a trailer, the method comprising:

[0036] The first acquisition module is used to acquire the first point cloud data, which is the point cloud data of the current vehicle in the current frame.

[0037] The second acquisition module is used to acquire historical straight line parameters and determine the search range based on the historical straight line parameters. The historical straight line parameters are the parameters of the edge line of the vehicle trailer in the current historical frame, which are determined based on the point cloud data of the current vehicle historical frame.

[0038] The determination module is used to determine the optimal straight line pair based on the search range, the width of the trailer, and the first point cloud data, and to determine the second point cloud data based on the optimal straight line pair;

[0039] The processing module is used to determine the target straight line based on the second point cloud data, and to determine the yaw angle based on the target straight line.

[0040] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following steps:

[0041] Acquire the first point cloud data, which is the point cloud data of the current vehicle in the current frame;

[0042] Obtain historical straight line parameters and determine the search range based on the historical straight line parameters. The historical straight line parameters are the parameters of the edge line of the vehicle trailer in the current historical frame, which are determined based on the point cloud data of the current vehicle historical frame.

[0043] Based on the search range, the width of the vehicle trailer, and the first point cloud data, determine the optimal straight line pair, and based on the optimal straight line pair, determine the second point cloud data;

[0044] The target straight line is determined based on the second point cloud data, and the yaw angle is determined based on the target straight line.

[0045] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, performs the following steps:

[0046] Acquire the first point cloud data, which is the point cloud data of the current vehicle in the current frame;

[0047] Obtain historical straight line parameters and determine the search range based on the historical straight line parameters. The historical straight line parameters are the parameters of the edge line of the vehicle trailer in the current historical frame, which are determined based on the point cloud data of the current vehicle historical frame.

[0048] Based on the search range, the width of the vehicle trailer, and the first point cloud data, determine the optimal straight line pair, and based on the optimal straight line pair, determine the second point cloud data;

[0049] The target straight line is determined based on the second point cloud data, and the yaw angle is determined based on the target straight line.

[0050] The aforementioned method, apparatus, computer equipment, and storage medium for determining the yaw angle of a trailer-mounted vehicle involve: acquiring first point cloud data (the point cloud data of the current vehicle in the current frame); acquiring historical straight line parameters and determining a search range based on these parameters (the parameters of the trailer's edge lines in the current vehicle's historical frames, determined from the point cloud data of the current vehicle); determining an optimal pair of straight lines based on the search range, the width of the trailer, and the first point cloud data; determining second point cloud data based on the optimal pair of straight lines; determining a target straight line based on the second point cloud data; and determining the yaw angle based on the target straight line. This application filters the point cloud data by searching for straight lines. The search for straight lines is performed in pairs, which better reflects the structure of the trailer's side edge lines. The filtered point cloud is then fitted with straight lines to obtain the target straight line. Specifically, this application utilizes historical straight line parameters to determine the search range of the straight line, initially narrowing the search range and accelerating the determination of the straight line. Furthermore, by combining the width of the trailer with the search, this application simultaneously searches for a pair of edge lines on both sides of the trailer, effectively reducing interference from other noise points. Traditional point cloud filtering techniques rely on point coordinates and vehicle dimensions. This method lacks precision and can introduce significant noise. For instance, in the presence of radar reflection noise or linear noise from other vehicles, the point cloud may contain more than one straight line. Searching for only one line could result in the retention of side point clouds from other vehicles, causing interference. Attached Figure Description

[0051] Figure 1 This is an application environment diagram of a method for determining the yaw angle of a vehicle trailer in one embodiment;

[0052] Figure 2 This is a flowchart illustrating a method for determining the yaw angle of a vehicle trailer in one embodiment.

[0053] Figure 3 This is a schematic diagram of the coordinate system where the first point cloud data of the current vehicle is located in one embodiment;

[0054] Figure 4 This is a flowchart illustrating the steps of determining the optimal straight line pair based on the search range, the width of the vehicle trailer, and the first point cloud data, and determining the second point cloud data based on the optimal straight line pair, in one embodiment.

[0055] Figure 5 This is a schematic diagram illustrating the process of filtering first point cloud data to obtain second point cloud data in one embodiment.

[0056] Figure 6 This is a schematic diagram of a graph neural network model filtering the first point cloud data in one embodiment;

[0057] Figure 7This is a schematic diagram illustrating the principle of determining the quality of a straight line in one embodiment;

[0058] Figure 8 This is a schematic diagram of the structure of the optimal line pair in one embodiment;

[0059] Figure 9 This is a flowchart illustrating the steps of determining a target straight line based on second point cloud data and determining the yaw angle based on the target straight line in one embodiment.

[0060] Figure 10 This is a flowchart illustrating the method for determining the yaw angle of a vehicle trailer in another embodiment;

[0061] Figure 11 This is a structural block diagram of a device for determining the trailer yaw angle in one embodiment;

[0062] Figure 12 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0064] Please refer to Figure 1 , Figure 1 This is a schematic diagram illustrating the application environment of a method for determining the yaw angle of a vehicle trailer, provided as an exemplary embodiment of this application. For example... Figure 1 As shown, the application environment includes a current vehicle 100, which includes a trailer 101. A lidar scanning device 102 is also installed on the current vehicle 100. During autonomous driving, the lidar scanning device 102 scans the current vehicle 100 in real time to obtain first point cloud data, which is the point cloud data of the current frame of the current vehicle. The current vehicle 100 is equipped with a smart terminal, which can acquire the first point cloud data collected by the lidar scanning device 102, obtain historical straight line parameters, and determine the search range based on the historical straight line parameters. The historical straight line parameters are the parameters of the edge line of the trailer in the current vehicle's historical frame, determined based on the point cloud data of the current vehicle's historical frame. Based on the search range, the width of the trailer 101, and the first point cloud data, an optimal straight line pair is determined, and based on the optimal straight line pair, second point cloud data is determined. A target straight line is determined based on the second point cloud data, and the yaw angle is determined based on the target straight line, thus realizing the trailer yaw angle determination method of this application.

[0065] In one embodiment, such as Figure 2 As shown, a method for determining the yaw angle of a vehicle trailer is provided, which is then applied to... Figure 1 Taking a smart terminal as an example, the explanation includes the following steps:

[0066] S11. Obtain the first point cloud data, which is the point cloud data of the current vehicle in the current frame.

[0067] In this application, the aforementioned first point cloud data refers to the point cloud data of the current vehicle in the current frame, obtained in real time by a LiDAR scanning device installed on the vehicle during the autonomous driving process. The LiDAR scanning device can be one or more. In one embodiment, one LiDAR scanning device can be installed next to each of the left and right rearview mirrors of the vehicle.

[0068] Specifically, the first point cloud data includes the coordinates, color information, reflection intensity information, and plane normal vectors of each point in the current frame of the current vehicle.

[0069] During the current autonomous driving process, the LiDAR scanning device scans the vehicle in real time to obtain the first point cloud data, which is then sent to the intelligent terminal for processing. The LiDAR scanning device in this application can be a radar sensor, which can be one sensor, or two sensors (one on the left and one on the right) simultaneously collecting point cloud data from both parts, or any other number of sensors; no specific limitation is made here.

[0070] S12. Obtain historical straight line parameters and determine the search range based on the historical straight line parameters. The historical straight line parameters are the parameters of the edge line of the vehicle trailer in the current historical frame, which are determined based on the point cloud data of the current vehicle historical frame.

[0071] like Figure 3 As shown, Figure 3 This is a schematic diagram of the coordinate system where the first point cloud data of the current vehicle is located in one embodiment. Figure 3 In this application, the current vehicle 30 includes a front end 31 and a trailer 32. The point cloud data obtained by the lidar scanning device consists of points in three-dimensional space, with each point having three-dimensional coordinates. The pose of the vehicle's front end can be provided by hardware; all point cloud data in this application resides in a local coordinate system centered on the front end. Figure 3 In this application, the first point cloud data is located in a three-dimensional coordinate system with the central axis of the vehicle head 31 as the Y-axis, the center point of the vehicle head 31 as the origin, the line perpendicular to the Y-axis as the X-axis, and the line perpendicular to the XY plane as the Z-axis (not shown in the figure).

[0072] In this application, the aforementioned current vehicle may include a vehicle front and a trailer. The trailer includes edge lines on both sides. This application scans the current vehicle in real time to obtain point cloud data for each frame, and processes the point cloud data of each frame to obtain the line parameters corresponding to the point cloud data of each frame. Among them, the historical line parameters are the line parameters corresponding to the historical frames of the current vehicle. The line is the edge line on both sides of the trailer of the current vehicle. The historical line parameters may specifically include the angle between the line and the Y-axis and the intercept of the line on the X-axis. In another embodiment, the historical line parameters may also be the slope of the line and the angle between the line and the X-axis, or other parameters that can determine a line, which are not specifically limited here.

[0073] In this application, the search range can be the range of yaw angles and the range of the intercept of the straight line on the X-axis. The yaw angle is the angle between the straight line and the Y-axis. Assuming the current frame is frame T, the historical frames mentioned above can be frame T-1, frame T-2, or frame TN, etc. Multiple frames are also possible, without specific limitation. Generally, frame T-1 is taken as the historical frame to determine the search range.

[0074] When determining the search range based on historical line parameters, this application can determine the range of the angle between the line and the Y-axis and the range of the intercept of the line on the X-axis in the current frame based on the angle between the line and the Y-axis and the intercept of the line on the X-axis in the historical line parameters.

[0075] S13. Based on the search range, the width of the vehicle trailer, and the first point cloud data, determine the optimal straight line pair, and based on the optimal straight line pair, determine the second point cloud data.

[0076] In this application, the aforementioned optimal straight line pair is a pair of straight lines predicted based on the search range, the width of the vehicle trailer, and the first point cloud data to predict the edge lines on both sides of the vehicle trailer.

[0077] In this application, the filtering of the first point cloud data is achieved by searching for straight lines. This search is performed in pairs, meaning that the search considers the width of the trailer while simultaneously searching for a pair of straight lines on either side of the trailer. This makes the algorithm more robust and the filtering more accurate. Specifically, based on the search range, the range of straight line parameters for the edge lines on both sides of the trailer corresponding to the current frame can be obtained. Based on this range, multiple straight lines can be identified. Further filtering based on the trailer width allows for the selection of multiple pairs of straight lines, resulting in the optimal pair. Points that are too far from this optimal pair are then filtered out, yielding the second point cloud data.

[0078] S14. Determine the target straight line based on the second point cloud data, and determine the yaw angle based on the target straight line.

[0079] In this application, the target straight line is the edge line of one side of the trailer. The yaw angle can be obtained from the angle between the target straight line and the Y-axis.

[0080] The target straight line can be determined by fitting the coordinate values ​​of the second point cloud data to a straight line.

[0081] In one embodiment, the aforementioned historical frame includes the previous frame, and the historical straight line parameters include the parameters of the target edge lines among the two side edge lines of the trailer in the previous frame. Determining the search range based on the historical straight line parameters includes:

[0082] Based on the parameters of the target edge line of the vehicle in the previous frame and the preset threshold, the range of parameters of the target edge line in the current frame is determined, and the search range is obtained.

[0083] In this application, it is assumed that the aforementioned historical straight line parameters are the angle between the target edge line of the vehicle in the previous frame and the Y-axis, and the intercept on the X-axis. Assuming the current frame is the frame at time T, and the historical frame is the frame corresponding to time T-1, the historical straight line parameters corresponding to time T-1 are an angle of 30 degrees between the line and the Y-axis, and an intercept of 1 meter on the X-axis. Based on these historical straight line parameters, the search range can be determined as the range of the angle between the target edge line at time T and the Y-axis [30-Δθ, 30+Δθ], where Δθ is a settable parameter. The range of the intercept of the target edge line on the X-axis in the current frame is [1-ΔX, 1+ΔX] meters, where ΔX is a settable parameter.

[0084] It should be noted that in special cases, such as when T=0, there will be no results at time T-1. In this case, we use the maximum search range, i.e., the yaw angle range is [-90, 90] degrees, and the target line intercept range on the X-axis is [1-ΔW, 1+ΔW] meters, where ΔW is a configurable parameter. Furthermore, if due to special circumstances such as occlusion, there are no results at time T-1, but results are found at time T-Δt (Δt is a configurable parameter, such as 2), we use the parameter corresponding to the target edge at time T-Δt to determine the search range and dynamically expand the search range. For example, we can simply expand Δθ and ΔX to Δt times their original values.

[0085] In this application, this implementation method can combine information from historical moments to narrow the search scope and accelerate the search process.

[0086] In one embodiment, such as Figure 4 As shown, the process of determining the optimal line pair based on the search range, the width of the vehicle trailer, and the first point cloud data, and then determining the second point cloud data based on the optimal line pair, can include:

[0087] S41. Project the three-dimensional point cloud corresponding to the first point cloud data onto a two-dimensional plane to obtain the projected point cloud;

[0088] S42. Determine the parameters of each predicted line according to the search range and the preset search interval;

[0089] S43. Based on the predicted straight line parameters and the current vehicle trailer width, determine multiple straight line pairs;

[0090] S44. Determine the mass of each straight line pair based on the points in the projected point cloud that are within the first preset range of distance to each straight line pair;

[0091] S45. Determine the optimal pair of lines based on the quality of each pair of lines;

[0092] S46. Determine the points within the second preset range of the optimal straight line pair based on the projected point cloud to obtain the second point cloud data.

[0093] In this application, the aforementioned projection of the three-dimensional point cloud corresponding to the first point cloud data onto a two-dimensional plane to obtain the projected point cloud may include:

[0094] Project each point of the first point cloud data onto the XY plane to obtain the projected point clouds on the XY plane. Specifically, discard the x, y, and z coordinates of each point to obtain the projected point clouds of each point on the XY plane, where each projected point contains (x, y) coordinates.

[0095] The search intervals mentioned above are preset values. In this application, for example, the search interval for the yaw angle is set to 10 degrees, and the search interval for the intercepts of the two edge lines on the X-axis is 0.5 meters. When the search range is a yaw angle of [20, 40] degrees, and the intercept of one of the two edge lines on the X-axis is [0.5, 1.5] meters, then the combinations of the predicted straight line parameters can be obtained as follows:

[0096] (20 degrees, 0.5 meters), (30 degrees, 0.5 meters), (40 degrees, 0.5 meters), (20 degrees, 1.0 meter), (30 degrees, 1.0 meter), (40 degrees, 1.0 meter), (20 degrees, 1.5 meters), (30 degrees, 1.5 meters) and (40 degrees, 1.5 meters).

[0097] Furthermore, for each combination of predicted straight line parameters, a corresponding straight line can be drawn in the point cloud, and then, based on the width of the trailer, another straight line can be drawn to match each straight line, thereby determining multiple pairs of straight lines.

[0098] Furthermore, the aforementioned determination of the quality of each line pair based on points in the projected point cloud that are within a first preset range from each line pair may include:

[0099] Obtain points in the projected point cloud that are within a first preset range from each pair of lines, and evaluate the quality of each pair of lines based on the obtained points.

[0100] The pair of lines with the best quality is selected as the optimal pair of lines mentioned above.

[0101] The quality of each straight line pair can be evaluated based on the number of points surrounding the line or on the effective line length. The first preset range mentioned above can be set according to actual needs.

[0102] The above-mentioned process of determining the optimal distance line pair from the projected point cloud to points within a second preset range, and obtaining the second point cloud data, may include:

[0103] Based on the distance of each projection point to the optimal line pair, points outside the second preset range of the distance to the optimal line pair are filtered out to obtain the second point cloud data.

[0104] The second preset range mentioned above can be set according to actual needs.

[0105] Please refer to Figure 5 , Figure 5 This is a schematic diagram illustrating the process of filtering first point cloud data to obtain second point cloud data in one embodiment. Figure 5 The data includes the current vehicle 80, the front of the vehicle 81, and the trailer 82. It also includes points 83 (circled) retained after filtering the first point cloud data, points 84 (rectangled) that were filtered out, and the optimal straight line pair 85. Figure 5 It can be seen that the retained points 83 are within the second preset range of the optimal straight line pair 85, thus forming the second point cloud data.

[0106] In this application, the first point cloud data is filtered by searching for straight lines. Utilizing the width of the vehicle trailer, a pair of straight lines are searched simultaneously. When radar reflection noise or noise from other vehicles' straight lines is present, more than two straight lines may exist in the point cloud. In this case, searching only one straight line might result in the retention of side point clouds of other vehicles, causing interference. This application also incorporates historical information to narrow the search range and accelerate the search speed.

[0107] In one embodiment, the above-mentioned process of projecting the three-dimensional point cloud corresponding to the first point cloud data onto a two-dimensional plane to obtain the projected point cloud may include:

[0108] Based on the first point cloud data and the pre-trained graph neural network model, the third point cloud data is obtained. The graph neural network model is used to perform non-linear transformation on the first point cloud data to obtain the category label value of each point, so as to filter each point according to the category label value of each point to obtain the third point cloud data.

[0109] The 3D point cloud corresponding to the third point cloud data is projected onto a 2D plane to obtain the projected point cloud.

[0110] In this application, the aforementioned graph neural network model can be a general term for a class of algorithms that can be used to process point clouds. Generally, any network capable of performing point cloud semantic segmentation tasks can be trained as this graph neural network model. For example, common graph neural network models include PointNet, PointNet++, and PointTransformer.

[0111] The process of obtaining the third point cloud data based on the first point cloud data and a pre-trained graph neural network model can include:

[0112] The first point cloud data is processed using a graph neural network model to obtain the category label value corresponding to each point;

[0113] Based on the category label values ​​corresponding to each point, the sigmoid function is used to scale the category label values ​​to the (0, 1) interval;

[0114] The scaled values ​​are compared with a preset threshold. Points with scaled values ​​greater than the preset threshold are identified as noise points. All noise points are deleted to obtain the third point cloud data.

[0115] The sigmoid function, also known as the logistic function, is used for the output of hidden layer neurons. Its value range is (0,1). It can map a real number to the interval (0,1) and can be used for binary classification.

[0116] Specifically, the first point cloud data is input into a graph neural network model. Assume it includes M1 points, each with three-dimensional coordinates (x, y, z). Through multiple nonlinear transformations of the graph neural network model, each point can be mapped to a one-dimensional space, obtaining the predicted class label value for each point. To obtain the probability that each point belongs to noise, the sigmoid function is used to scale the class label value to the (0, 1) interval. The scaled value is compared with a preset threshold; if the probability is greater than the preset threshold (e.g., 0.5), the point is considered noise.

[0117] Please refer to Figure 6 , Figure 6 This is a schematic diagram illustrating how a graph neural network model filters first point cloud data in one embodiment. Figure 6 In the present vehicle 50, there is a cab 51 and a trailer 52. Figure 5 In the graph, the circle represents point 53, which is the point that remains after filtering, and the rectangle represents point 54, which is the point that was filtered out by the graph neural network.

[0118] In this application, the aforementioned graph neural network model encodes the semantic information of each point by calculating category label values, including local geometric structure (such as topology, curvature, and density) and global context (such as the symmetry of points on the left and right sides of the trailer). This helps distinguish radar points on the side plane of the trailer from radar noise on rainwater, tires, and leaves. Noise around the trailer can affect the determination of candidate target lines, thus increasing the error in yaw angle prediction. Traditional techniques do not utilize graph neural network models to filter point cloud data, resulting in the loss of the three-dimensional structure of the point cloud and making it difficult to filter out noise with special geometric semantics.

[0119] In one embodiment, the above method may further include:

[0120] Based on the coordinates of each point in the first point cloud data, the current vehicle pose, and the current vehicle size information, the first point cloud data is filtered to obtain the fourth point cloud data.

[0121] Based on the first point cloud data and the pre-trained graph neural network model, the third point cloud data is obtained, including:

[0122] The third point cloud data is obtained based on the fourth point cloud data and the pre-trained graph neural network model.

[0123] In this application, the current vehicle pose can include the current vehicle's position and rotation. The current vehicle's size information can include its length, width, and height. Before inputting the first point cloud data into the graph neural network model, a filtering operation can be performed on the first point cloud data. Specifically, this application filters out points that are not within the trailer height range based on the z-coordinate values ​​of each point. Points that are too far from the trailer are filtered out based on the x and y coordinate values ​​of each point, the dimensions of the tractor and trailer, and the resulting third point cloud data is then input into the graph neural network model for processing.

[0124] This application can perform preliminary filtering of the first point cloud data through this implementation method, and filter out points that obviously do not belong to the current vehicle based on the coordinate values ​​of each point and the size of the current vehicle. This application makes the subsequent data processing results more accurate through multiple filtering.

[0125] In one embodiment, determining the quality of each line pair based on points in the projected point cloud that are within a first preset range from each line pair may include:

[0126] Project the projection point cloud of each line pair within the first preset range onto the corresponding line pair to obtain each target projection point on each line pair.

[0127] The mass of each straight line pair is determined based on the distance between the target projection points on each straight line pair.

[0128] Please refer to Figure 7 , Figure 7 This is a schematic diagram illustrating the determination of straight-line quality in one embodiment. Figure 7 The diagram includes one straight line 61 from each pair of straight lines, points 62 (symbolized by triangles) and 63 (symbolized by circles) surrounding the straight line. Point 62 is located outside the first preset range, and point 63 is located within the first preset range. In this application, the estimation of straight line quality is as follows: Figure 7 As shown, points 63 within the first preset range of distance to each line are used to evaluate the quality of the lines. All points 63 within the first preset range of distance to each line are projected onto the corresponding lines along the direction perpendicular to the lines, resulting in target projection points 64. 64 specifically includes points A, B, C, D, E, and F, where BC, CD, and DE have equal lengths. Further, a circle with a fixed radius is drawn with each target projection point 64 as its center and a preset value as its radius. All points within the circular region 65 are considered neighbors of each target projection point, and the line connecting the center of the circle to its corresponding neighbor is considered an edge. The lengths of all edges are added together, while the lengths of overlapping edges are removed, to obtain the effective length of each line. The effective length of each line is added to the effective length of its paired line to obtain the effective length of each pair of lines.

[0129] It should be noted that if a point is an isolated point, meaning there are no other points within a circular area centered at that point and with a preset value as the radius, then a pre-set fixed parameter must be added when calculating the effective length for such an isolated point. (See above.) Figure 7 In this equation, the effective length of the line is 2e + BE, where e is a pre-defined fixed parameter.

[0130] In this application, during each search, the mass of a pair of lines is calculated simultaneously and added together as the total effective length of that search. Then, the pair of lines with the longest effective length is selected as the optimal pair of lines.

[0131] In this application, when calculating the quality of each straight line, the structure of the straight line is explicitly modeled, and the points around each straight line are projected onto the straight line to calculate the effective length, which can reduce the influence of dense noise.

[0132] Please refer to Figure 8 , Figure 8 This is a schematic diagram of the structure of the optimal line pair in one embodiment. Figure 8 The present invention includes line pair 71 and line pair 72. Line pair 71 is the optimal line pair obtained according to conventional technical solutions. Line pair 72 is the line pair obtained according to the technical solution of this application.

[0133] Traditional techniques (such as the Hough transform) typically use the number of points to evaluate the quality of a line, resulting in an optimal pair of lines, such as... Figure 7 The line pair 71 is shown. By calculating the quality of the line pair using the technical solution of this application, the optimal line pair 72 can be obtained. From... Figure 7 It can be seen that although line pair 71 contains more points, its effective length is less than that of line pair 72. Therefore, according to the technical solution of this application, it can be concluded that the quality of line pair 72 is significantly better than that of line pair 71.

[0134] In one embodiment, such as Figure 9 As shown, the above-mentioned process of determining the target straight line based on the second point cloud data and determining the yaw angle based on the target straight line can include:

[0135] S91. Perform a preset number of iterations of sampling on the second point cloud data according to the random sampling consensus algorithm.

[0136] S92. Perform line fitting on the extracted point cloud data according to the least squares method to determine the line corresponding to each iteration;

[0137] S93. Determine the quality of the line corresponding to each iteration based on the point cloud data extracted in each iteration, and determine the target line based on the quality of the line corresponding to each iteration.

[0138] S94. Obtain the yaw angle based on the angle between the target straight line and the centerline of the current vehicle's front.

[0139] In this application, the target straight line mentioned above is one of the straight lines on both sides of the vehicle trailer. After obtaining the second point cloud data of the optimal straight line, this application processes the second point cloud data using RANSAC (Random Sample Consensus Algorithm). The RANSAC algorithm iteratively filters out noise points on both sides through subset sampling.

[0140] For example, suppose in the second point cloud data, L points are within the second threshold range of the distance to the left edge of the trailer, and R points are within the second threshold range of the distance to the right edge of the trailer. We perform RANSAC filtering on both sides. For the L points on the left, we first randomly select L1-1 points and use the least squares method to fit a straight line to the point cloud data of these L1-1 points, obtaining a straight line. We then calculate the quality M1 of this straight line, as described above, and will not elaborate further here. It is worth noting that, to make the RANSAC filtering more refined, the threshold for determining the points around the straight line when calculating the line quality is usually lower than the value set in the previous line search step. Then, we use this straight line to filter out points that are too far from the straight line, as described above, leaving L1-2 points. Next, we repeat the above steps using the remaining L1-2 points. The repeated steps include fitting a straight line using the least squares method, calculating the line quality, and filtering out points closer to the new straight line from the L points. We stop repeating the sequence when the quality of the line obtained after a repetition is worse than the previous one or when the maximum number of repetitions is exceeded, thus completing the current iteration. The number of repetitions is a configurable parameter.

[0141] After completing the first full iteration, randomly select L2-1 points from the L points and repeat the iteration.

[0142] By analogy, we can obtain the final line quality M1, M2, M3...Mn after each sampling iteration. Based on the quality of the lines obtained after each sampling iteration, the line with the best quality is selected as the RANSAC result for the L points on the left. Similarly, we can obtain the RANSAC result for the R points on the right, and select the line with the best quality from the left and right RANSAC results to obtain the target line.

[0143] In this application, because points around the line are always retained during the process of determining the second point cloud data (i.e., the line search process), the distance threshold is set relatively large to consider time efficiency, which may result in some noise. If the least squares method is used directly, the final estimated result will have a small bias. Therefore, the RANSAC algorithm can be used for refinement to reduce bias.

[0144] like Figure 10 As shown, in one possible embodiment of this application, the method for determining the yaw angle of the vehicle-mounted trailer described above may include the following steps:

[0145] S101, Obtain the first point cloud data;

[0146] S102. Based on the coordinate values ​​of each point in the first point cloud data, the current vehicle pose, and the current vehicle size information, filter the first point cloud data to obtain the fourth point cloud data.

[0147] S103. Based on the fourth point cloud data and the pre-trained graph neural network model, obtain the third point cloud data;

[0148] S104. Project the three-dimensional point cloud corresponding to the third point cloud data onto a two-dimensional plane to obtain the projected point cloud;

[0149] S105. Obtain historical line parameters and determine the search range based on the historical line parameters;

[0150] S106. Based on the search range, the width of the trailer, and the projected point cloud, determine the optimal straight line pair, and based on the optimal straight line pair and the projected point cloud, determine the second point cloud data.

[0151] S107. The second point cloud data is processed using the RANSAC algorithm and the least squares method to determine the target straight line, and the yaw angle is determined based on the target straight line.

[0152] In this application, common dense noise points include rainwater noise, leaf noise, and radar reflection noise. When these noise points are close to the side of the trailer or have a planar structure, the aforementioned graph neural network model may have difficulty filtering them. Furthermore, the RANSAC algorithm, due to its randomness, may not be able to cover all areas and is less stable than the aforementioned graph neural network model. Therefore, adding steps S94 to S96, a line search step, between the aforementioned graph neural network model and the RANSAC algorithm can increase the stability of the RANSAC algorithm's search, i.e., reduce the upper limit of possible errors caused by the randomness of RANSAC. In addition, this application utilizes the trailer width and the effective length of the line to further reduce the impact of noise.

[0153] In one embodiment, such as Figure 11 As shown, a device for determining the yaw angle of a trailer is provided, comprising: a first acquisition module 11, a second acquisition module 12, a determination module 13, and a processing module, wherein:

[0154] The first acquisition module 11 is used to acquire the first point cloud data, which is the point cloud data of the current vehicle in the current frame;

[0155] The second acquisition module 12 is used to acquire historical straight line parameters and determine the search range based on the historical straight line parameters. The historical straight line parameters are the parameters of the edge line of the vehicle trailer in the current historical frame, which are determined based on the point cloud data of the current vehicle historical frame.

[0156] The determination module 13 is used to determine the optimal straight line pair based on the search range, the width of the trailer, and the first point cloud data, and to determine the second point cloud data based on the optimal straight line pair.

[0157] Processing module 14 is used to determine the target straight line based on the second point cloud data, and to determine the yaw angle based on the target straight line.

[0158] In one embodiment, the aforementioned historical frame includes the previous frame, and the aforementioned historical straight line parameters include the parameters of the target edge lines on both sides of the vehicle in the previous frame. The aforementioned second acquisition module 12 can determine the range of the target edge line parameters in the current frame based on the parameters of the target edge lines of the vehicle in the previous frame and a preset threshold, thereby obtaining the search range.

[0159] In one embodiment, the determining module 13 can project the three-dimensional point cloud corresponding to the first point cloud data onto a two-dimensional plane to obtain a projected point cloud. Based on the search range and a preset search interval, it determines the parameters of each predicted line. Based on the parameters of each predicted line and the width of the current vehicle's trailer, it determines multiple pairs of line pairs. Based on the points in the projected point cloud that are within a first preset range from each line pair, it determines the quality of each line pair. Based on the quality of each line pair, it determines the optimal line pair. Based on the projected point cloud, it determines the points within a second preset range from the optimal line pair to obtain the second point cloud data.

[0160] In one embodiment, the determination module 13 described above can also obtain third point cloud data based on the first point cloud data and the pre-trained graph neural network model. The graph neural network model is used to perform nonlinear transformation on the first point cloud data to obtain the category label value of each point, so as to filter each point according to the category label value of each point to obtain the third point cloud data. The three-dimensional point cloud corresponding to the third point cloud data is projected onto a two-dimensional plane to obtain the projected point cloud.

[0161] In one embodiment, the above-mentioned trailer yaw angle determination device further includes a filtering module (not shown). The filtering module can filter the first point cloud data according to the coordinate values ​​of each point in the first point cloud data, the pose of the current vehicle, and the size information of the current vehicle to obtain the fourth point cloud data. The determination module 13 can also obtain the third point cloud data according to the fourth point cloud data and the pre-trained graph neural network model.

[0162] In one embodiment, the determination module 13 described above can also project the projection point cloud of each line pair within a first preset range onto the corresponding line pair to obtain each target projection point on each line pair, and determine the quality of each line pair based on the distance between each target projection point on each line pair.

[0163] In one embodiment, the processing module 14 can perform a preset number of iterations of sampling on the second point cloud data according to the random sampling consensus algorithm, perform line fitting on the point cloud data sampled in each iteration according to the least squares method, determine the line corresponding to each iteration, determine the quality of the line corresponding to each iteration according to the point cloud data sampled in each iteration, determine the target line according to the quality of the line corresponding to each iteration, and obtain the yaw angle according to the angle between the target line and the centerline of the current vehicle's front.

[0164] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 12 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data such as the operating data of smart home devices. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a resource allocation method for compiling a virtual machine.

[0165] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following steps: acquiring first point cloud data, the first point cloud data being point cloud data of the current vehicle in the current frame; acquiring historical straight line parameters, and determining a search range based on the historical straight line parameters, the historical straight line parameters being parameters of the edge line of the trailer in the current vehicle's historical frame determined based on the point cloud data of the current vehicle's historical frame; determining an optimal straight line pair based on the search range, the width of the trailer, and the first point cloud data, and determining second point cloud data based on the optimal straight line pair; determining a target straight line based on the second point cloud data, and determining a yaw angle based on the target straight line.

[0166] In one embodiment, the aforementioned historical frame includes the previous frame, and the aforementioned historical straight line parameters include the parameters of the target edge lines among the two side edge lines of the trailer in the previous frame. When the processor executes the computer program to implement the aforementioned step of determining the search range based on the historical straight line parameters, the following steps are specifically implemented:

[0167] Based on the parameters of the target edge line of the vehicle in the previous frame and the preset threshold, the range of parameters of the target edge line in the current frame is determined, and the search range is obtained.

[0168] In one embodiment, when the processor executes the computer program to determine the optimal line pair based on the search range, the width of the trailer, and the first point cloud data, and to determine the second point cloud data based on the optimal line pair, the following steps are specifically implemented:

[0169] Project the 3D point cloud corresponding to the first point cloud data onto a 2D plane to obtain the projected point cloud;

[0170] Based on the search range and the preset search interval, determine the parameters of each predicted line;

[0171] Based on the predicted straight line parameters and the current vehicle's trailer width, multiple straight line pairs are determined.

[0172] The mass of each straight line pair is determined based on the points in the projected point cloud that are within a first preset range from each straight line pair.

[0173] Determine the optimal line pair based on the quality of each line pair;

[0174] Based on the projected point cloud, the point within the second preset range of the optimal straight line pair is determined, and the second point cloud data is obtained.

[0175] In one embodiment, when the processor executes a computer program to implement the above-described step of projecting the three-dimensional point cloud corresponding to the first point cloud data onto a two-dimensional plane to obtain the projected point cloud, the following steps are specifically implemented:

[0176] Based on the first point cloud data and the pre-trained graph neural network model, the third point cloud data is obtained. The graph neural network model is used to perform non-linear transformation on the first point cloud data to obtain the category label value of each point, so as to filter each point according to the category label value of each point to obtain the third point cloud data.

[0177] The 3D point cloud corresponding to the third point cloud data is projected onto a 2D plane to obtain the projected point cloud.

[0178] In one embodiment, when the processor executes a computer program, it further implements the following steps:

[0179] Based on the coordinates of each point in the first point cloud data, the current vehicle pose, and the current vehicle size information, the first point cloud data is filtered to obtain the fourth point cloud data.

[0180] When the processor executes the computer program to obtain the third point cloud data based on the first point cloud data and the pre-trained graph neural network model, the specific steps are as follows:

[0181] The third point cloud data is obtained based on the fourth point cloud data and the pre-trained graph neural network model.

[0182] In one embodiment, when the processor executes a computer program to determine the quality of each straight line pair based on points within a first preset range in the projected point cloud, the following steps are specifically implemented:

[0183] Project the projection point cloud of each line pair within the first preset range onto the corresponding line pair to obtain each target projection point on each line pair.

[0184] The mass of each straight line pair is determined based on the distance between the target projection points on each straight line pair.

[0185] In one embodiment, when the processor executes the computer program to implement the above-described steps of determining the target straight line based on the second point cloud data and determining the yaw angle based on the target straight line, the following steps are specifically implemented:

[0186] The second point cloud data is sampled a preset number of times according to the random sampling consensus algorithm.

[0187] The point cloud data extracted in each iteration is fitted with a straight line using the least squares method to determine the straight line corresponding to each iteration.

[0188] The quality of the line corresponding to each iteration is determined based on the point cloud data extracted in each iteration, and the target line is determined based on the quality of the line corresponding to each iteration.

[0189] The yaw angle is obtained by the angle between the target straight line and the centerline of the current vehicle's front end.

[0190] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon. When executed by a processor, the computer program performs the following steps: acquiring first point cloud data, the first point cloud data being the point cloud data of the current vehicle in the current frame; acquiring historical straight line parameters and determining a search range based on the historical straight line parameters, the historical straight line parameters being the parameters of the edge line of the trailer in the current vehicle's historical frame determined based on the point cloud data of the current vehicle's historical frame; determining an optimal straight line pair based on the search range, the width of the trailer, and the first point cloud data, and determining second point cloud data based on the optimal straight line pair; determining a target straight line based on the second point cloud data, and determining a yaw angle based on the target straight line.

[0191] In one embodiment, the aforementioned historical frame includes the previous frame, and the aforementioned historical straight line parameters include the parameters of the target edge lines among the two side edge lines of the trailer in the previous frame. When the computer program is executed by the processor to implement the aforementioned step of determining the search range based on the historical straight line parameters, the following steps are specifically implemented:

[0192] Based on the parameters of the target edge line of the vehicle in the previous frame and the preset threshold, the range of parameters of the target edge line in the current frame is determined, and the search range is obtained.

[0193] In one embodiment, when a computer program is executed by a processor to perform the steps described above—determining the optimal line pair based on the search range, the width of the trailer, and the first point cloud data, and determining the second point cloud data based on the optimal line pair—the following steps are specifically implemented:

[0194] Project the 3D point cloud corresponding to the first point cloud data onto a 2D plane to obtain the projected point cloud;

[0195] Based on the search range and the preset search interval, determine the parameters of each predicted line;

[0196] Based on the predicted straight line parameters and the current vehicle's trailer width, multiple straight line pairs are determined.

[0197] The mass of each straight line pair is determined based on the points in the projected point cloud that are within a first preset range from each straight line pair.

[0198] Determine the optimal line pair based on the quality of each line pair;

[0199] Based on the projected point cloud, the point within the second preset range of the optimal straight line pair is determined, and the second point cloud data is obtained.

[0200] In one embodiment, when a computer program is executed by a processor to implement the above-described step of projecting the three-dimensional point cloud corresponding to the first point cloud data onto a two-dimensional plane to obtain the projected point cloud, the following steps are specifically implemented:

[0201] Based on the first point cloud data and the pre-trained graph neural network model, the third point cloud data is obtained. The graph neural network model is used to perform non-linear transformation on the first point cloud data to obtain the category label value of each point, so as to filter each point according to the category label value of each point to obtain the third point cloud data.

[0202] The 3D point cloud corresponding to the third point cloud data is projected onto a 2D plane to obtain the projected point cloud.

[0203] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented:

[0204] Based on the coordinates of each point in the first point cloud data, the current vehicle pose, and the current vehicle size information, the first point cloud data is filtered to obtain the fourth point cloud data.

[0205] When the computer program is executed by the processor to obtain the third point cloud data based on the first point cloud data and the pre-trained graph neural network model, the specific steps are as follows:

[0206] The third point cloud data is obtained based on the fourth point cloud data and the pre-trained graph neural network model.

[0207] In one embodiment, when a computer program is executed by a processor to implement the above-described step of determining the quality of each line pair based on points within a first preset range of distance from each line pair in the projected point cloud, the following steps are specifically implemented:

[0208] Project the projection point cloud of each line pair within the first preset range onto the corresponding line pair to obtain each target projection point on each line pair.

[0209] The mass of each straight line pair is determined based on the distance between the target projection points on each straight line pair.

[0210] In one embodiment, when a computer program is executed by a processor to perform the steps described above—determining the target straight line based on the second point cloud data and determining the yaw angle based on the target straight line—the following steps are specifically implemented:

[0211] The second point cloud data is sampled a preset number of times according to the random sampling consensus algorithm.

[0212] The point cloud data extracted in each iteration is fitted with a straight line using the least squares method to determine the straight line corresponding to each iteration.

[0213] The quality of the line corresponding to each iteration is determined based on the point cloud data extracted in each iteration, and the target line is determined based on the quality of the line corresponding to each iteration.

[0214] The yaw angle is obtained by the angle between the target straight line and the centerline of the current vehicle's front end.

[0215] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0216] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0217] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for determining the yaw angle of a vehicle-mounted trailer, characterized in that, The method includes: Acquire the first point cloud data, which is the point cloud data of the current vehicle in the current frame; Obtain historical straight line parameters, which are the parameters of the edge line of the trailer corresponding to the historical frame determined based on the point cloud data of the current vehicle's historical frame. The historical straight line parameters include the parameters of the target edge line in the edge lines on both sides of the trailer in the previous frame of the current frame. Based on the parameters of the target edge line of the vehicle in the previous frame and a preset threshold, the range of the parameters of the target edge line in the current frame is determined to obtain the search range. Based on the search range, the width of the vehicle trailer, and the first point cloud data, determine the optimal straight line pair, and based on the optimal straight line pair, determine the second point cloud data; The target straight line is determined based on the second point cloud data, and the yaw angle is determined based on the target straight line.

2. The method according to claim 1, characterized in that, The step of determining the optimal line pair based on the search range, the width of the vehicle trailer, and the first point cloud data, and determining the second point cloud data based on the optimal line pair, includes: The three-dimensional point cloud corresponding to the first point cloud data is projected onto a two-dimensional plane to obtain the projected point cloud. Based on the search range and the preset search interval, determine the parameters of each predicted line; Based on the predicted straight line parameters and the current vehicle's trailer width, multiple pairs of straight lines are determined; The mass of each straight line pair is determined based on the points in the projected point cloud that are within a first preset range from each straight line pair. The optimal line pair is determined based on the quality of each line pair. Based on the projected point cloud, points within a second preset range that are far from the optimal straight line pair are determined to obtain the second point cloud data.

3. The method according to claim 2, characterized in that, The step of projecting the three-dimensional point cloud corresponding to the first point cloud data onto a two-dimensional plane to obtain the projected point cloud includes: Based on the first point cloud data and the pre-trained graph neural network model, the third point cloud data is obtained. The graph neural network model is used to perform nonlinear transformation on the first point cloud data to obtain the category label value of each point, so as to filter each point according to the category label value of each point to obtain the third point cloud data. The three-dimensional point cloud corresponding to the third point cloud data is projected onto a two-dimensional plane to obtain the projected point cloud.

4. The method according to claim 3, characterized in that, The method further includes: Based on the coordinate values ​​of each point in the first point cloud data, the pose of the current vehicle, and the size information of the current vehicle, the first point cloud data is filtered to obtain the fourth point cloud data. The process of obtaining the third point cloud data based on the first point cloud data and the pre-trained graph neural network model includes: The third point cloud data is obtained based on the fourth point cloud data and the pre-trained graph neural network model.

5. The method according to claim 2, characterized in that, The step of determining the quality of each line pair based on points in the projected point cloud that are within a first preset range from each line pair includes: Project the projection point cloud within the first preset range of each line pair onto the corresponding line pair to obtain each target projection point on each line pair. The mass of each line pair is determined based on the distance between the target projection points on each line pair.

6. The method according to claim 1, characterized in that, The step of determining the target straight line based on the second point cloud data and determining the yaw angle based on the target straight line includes: The second point cloud data is sampled a preset number of times according to the random sampling consensus algorithm; The point cloud data extracted in each iteration is fitted with a straight line using the least squares method to determine the straight line corresponding to each iteration. The quality of the line corresponding to each iteration is determined based on the point cloud data extracted in each iteration, and the target line is determined based on the quality of the line corresponding to each iteration. The yaw angle is obtained by the angle between the target straight line and the centerline of the current vehicle's front end.

7. A device for determining the yaw angle of a trailer, characterized in that, The device includes: The first acquisition module is used to acquire the first point cloud data, which is the point cloud data of the current vehicle in the current frame; The second acquisition module is used to acquire historical straight line parameters. The historical straight line parameters are the parameters of the edge line of the vehicle frame corresponding to the historical frame, which are determined based on the point cloud data of the historical frame of the current vehicle. The historical straight line parameters include the parameters of the target edge line in the two side edge lines of the vehicle frame in the previous frame of the current frame. Based on the parameters of the target edge line of the vehicle frame in the previous frame and a preset threshold, the range value of the parameters of the target edge line in the current frame is determined to obtain the search range. The determination module is used to determine the optimal straight line pair based on the search range, the width of the vehicle trailer, and the first point cloud data, and to determine the second point cloud data based on the optimal straight line pair; The processing module is used to determine the target straight line based on the second point cloud data, and to determine the yaw angle based on the target straight line.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images