A method and system for distinguishing and labeling moving and static targets in point cloud data

Abstract

The present invention discloses a method and system for distinguishing and labeling moving and static targets of point cloud data, which relates to the field of autonomous driving technology. The method for distinguishing and labeling moving and static targets is as follows: obtaining all 3D point cloud data to be labeled; converting the 3D point cloud data in the current frame point cloud coordinate system into 3D point cloud data in the world coordinate system; performing interpolation calculation on the converted 3D point cloud data; predicting subsequent interpolation based on the interpolation of various known 3D point cloud data targets. The system is used to execute the method. The present invention can realize the automatic interpretation of moving targets and static targets, greatly reduce the workload of subsequent labeling personnel, eliminate the manual judgment process, and overall improve the overall labeling efficiency of subsequent 3D point cloud targets.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving, and particularly relates to a method and system for distinguishing and labeling moving and static targets of point cloud data. Background Art

[0002] In the scenario of vehicle autonomous driving, on-vehicle lidar continuously collects external scene targets of the vehicle to form point cloud data. The 3D point cloud generated by lidar can be used for the training of the autonomous driving system after annotation. With the improvement of performance, the amount of training data required for autonomous driving technology almost grows exponentially. During the collection process, since the collected targets may be dynamically moving (such as other vehicles driving on the road) or may be stationary (such as road signs, traffic lights), it is necessary to distinguish moving targets from static targets during annotation, which helps to improve the perception, decision-making, and control capabilities of the autonomous driving system during training.

[0003] In the existing 3D point cloud data collected, the moving and static targets are generally distinguished based on the manual interpretation of annotators, and then specific annotation work is carried out. However, due to the limited capabilities of annotators, empirical mistakes may also occur, confusing the moving and static targets. For example, some stationary vehicles may be labeled as moving targets, which not only increases the workload of annotators but also results in low-quality annotation data due to incorrect annotation operations, affecting the training effect of the autonomous driving model.

[0004] In this case, there are generally the following solutions: for example, a correction method based on manual operation can be adopted. The operator identifies multiple frames of images of the captured target manually, and determines whether the target is a moving target or a static target according to the position of the target object, etc. However, this method not only has a low accuracy rate, but also greatly increases the workload of the annotators, resulting in an increase in labor costs. Another example is that the Chinese invention patent with the publication number CN116205973A discloses a method and system for labeling continuous frame data of laser point clouds. Its technical solution is as follows: collecting point cloud data, obtaining the pose information of the point cloud, and dividing the point cloud data into multiple task packages according to a preset frame; for a single task package, adjusting the pose information of the first frame of point cloud in the task package; dividing the object to be labeled into static objects and dynamic objects for labeling; it sets a speed range to distinguish dynamic targets and static targets through a preset setting, and cannot achieve the effect of directly and automatically distinguishing dynamic and static targets. Moreover, this solution still has the risk of labeling delicate targets as dynamic targets, resulting in an impact on the accuracy of subsequent labeling of the target to be labeled. Another example is that the Chinese invention patent with the publication number CN117495912A discloses a method, device, equipment and storage medium for labeling dynamic objects in multiple frames of point clouds. Its technical solution is as follows: obtaining the positioning information of the vehicle; obtaining multiple frames of point cloud data of the vehicle; converting the multiple frames of point cloud data into multiple first world coordinate data according to the positioning information, and superimposing them to obtain multiple second world coordinate data; obtaining the first labeling information and the second labeling information of the dynamic object in the first frame and the last frame of the point cloud data; obtaining the position and direction of the dynamic object in the first frame and the last frame of the point cloud data; using an interpolation processing method to obtain the motion trajectory of the dynamic object according to the first labeling information, the second labeling information, the position and the direction; processing the motion trajectory to obtain the labeling information of the dynamic object; this solution only labels dynamic objects and cannot achieve the effect of automatically distinguishing dynamic and static objects before labeling.

[0005] Therefore, it is necessary to design a method and system for automatically distinguishing and labeling dynamic and static targets, which can solve the problems existing in the prior art, such as the large workload, low work efficiency and low labeling accuracy of manual labeling. At the same time, it can also solve the problem that it is impossible to conveniently and quickly automatically distinguish the dynamic and static states of the targets to be labeled before labeling in the prior art. Summary of the Invention

[0006] In view of the problems existing in the prior art, the present invention provides a method and system for distinguishing and labeling dynamic and static targets of point cloud data.

[0007] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] A method for distinguishing and labeling static and dynamic targets in point cloud data includes the following steps: S1. Obtain all 3D point cloud data to be labeled; S2. Convert the 3D point cloud data in the current frame point cloud coordinate system into 3D point cloud data in the world coordinate system; S3. Perform interpolation calculation on the converted 3D point cloud data. During the interpolation calculation process, set the target interpolation point P(tx, ty). If P(tx, ty) = 0, then determine that the target is a static target; if P(tx, ty) ≠ 0, then determine that the target is a dynamic target; S4. Based on the interpolation of various known 3D point cloud data targets, predict subsequent interpolation and labeling. Among them, the prediction method is: perform linear interpolation between two stereo frames in the point cloud data. The linear interpolation process considers three parameters: the size, orientation angle, and center point position of the stereo frame. By calculating the difference values corresponding to these three parameters respectively, predict the center point position, size, and orientation angle of the subsequent target, and perform automatic interpolation and labeling.

[0009] Based on the above technical solution, further, in S1, based on the visualization interface, obtain the 3D point cloud data of a certain target, and select any range of the target to be labeled from these 3D point cloud data.

[0010] Based on the above technical solution, further, the formation process of the visualization interface is as follows: According to the coordinates of any 3D point cloud data at the upper left corner and the coordinates of any 3D point cloud data at the lower right corner in the same frame within the range of the frame, form two-dimensional rectangular frames on the X-axis and Y-axis planes respectively, and traverse all 3D point cloud data within the range of the two-dimensional rectangular frame to determine the minimum value min_z and the maximum value max_z of all 3D point cloud data within the range of the two-dimensional rectangular frame in the Z-axis direction. Use the difference between the minimum value min_z and the maximum value max_z as the height of the three-dimensional rectangular frame, and generate the starting three-dimensional rectangular frame. This three-dimensional rectangular frame is the visualization interface.

[0011] Based on the above technical solution, further, the prediction process is as follows:

[0012] Step S41: Data preparation;

[0013] Step S42: Calculate the difference values of each parameter;

[0014] Step S43: Determine the interpolation steps;

[0015] Step S44: Perform linear interpolation;

[0016] Step S45: Generate interpolation targets;

[0017] Step S46: Output the interpolation results.

[0018] Based on the above technical solution, further, in step S41, the data are the parameters of the three-dimensional boxes A and B, and the parameters include the center point position, size, and orientation angle;

[0019] The center point positions are: A_center(x1, y1, z1), B_center(x2, y2, z2);

[0020] The sizes are: A_size(l1, w1, h1), B_size(l2, w2, h2);

[0021] The orientation angles are: A_orientation(θ1), B_orientation(θ2); where A_center(x1, y1, z1) is the center point position of the three-dimensional box A, B_center(x2, y2, z2) is the center point position of the three-dimensional box B; l1 is the length of the three-dimensional box A, w1 is the width of the three-dimensional box A, h1 is the height of the three-dimensional box A; l2 is the length of the three-dimensional box B, w2 is the width of the three-dimensional box B, h2 is the height of the three-dimensional box B; A _ orientation(θ1) is the orientation angle of the three-dimensional box A, B _ orientation(θ2) is the orientation angle of the three-dimensional box B.

[0022] Based on the above technical solution, further, in step S42, the process of calculating the difference values is as follows:

[0023] Calculate the difference values of the center point positions:

[0024] Δx = x2 - x1;

[0025] Δy = y2 - y1;

[0026] Δz = z2 - z1;

[0027] Calculate the difference values of the sizes:

[0028] Δl = l2 - l1;

[0029] Δw = w2 - w1;

[0030] Δh = h2 - h1;

[0031] Calculate the difference value of the orientation angle: Δθ = θ2 - θ1; where θ1 is the angle parameter of the three-dimensional box A, and θ2 is the angle parameter of the three-dimensional box B.

[0032] Based on the above technical solution, further, in step S43, according to the target number N to be interpolated, determine the interpolation step: step = 1 / (N + 1).

[0033] Based on the above technical solution, further, in step S44, for each interpolation target i, calculate the interpolation parameter, where i = 1, 2,..., N; the process is as follows:

[0034] The interpolation factor t = i × number of steps;

[0035] Interpolation of the center point position:

[0036] x i = x1 + t × Δx;

[0037] y i = y1 + t × Δy;

[0038] z i = z1 + t × Δz;

[0039] Interpolation of the size:

[0040] l i = l1 + t × Δl;

[0041] w i = w1 + t × Δw;

[0042] h i = h1 + t × Δh;

[0043] Interpolation of the orientation angle: θ i = θ1 + t × Δθ.

[0044] Based on the above technical solution, further, in step S45, according to the parameters calculated by interpolation, the stereo box parameters of the interpolation target are:

[0045] The center point position of the interpolation target i: (x_i, y_i, z_i);

[0046] The size of the interpolation target i: (l_i, w_i, h_i);

[0047] The orientation angle of the interpolation target i: θ_i.

[0048] A static and dynamic target discrimination annotation system for point cloud data, used to execute a static and dynamic target discrimination annotation method for point cloud data; includes an annotation module, a coordinate system conversion module, an interpolation calculation module, and an interpolation prediction module; the annotation module is used to start the tool for 3D point cloud data annotation; the coordinate system conversion module is used to convert the current frame point cloud coordinate system into the world coordinate system; the interpolation calculation module performs interpolation calculation on the 3D point cloud data of the same target in different frames loaded based on the world coordinate system; the interpolation prediction module predicts the interpolation of the moving target in the subsequent time series and performs the annotation action for the moving target.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] (1) The present invention can achieve the automatic distinction between moving targets and static targets. Specifically, in the original autonomous driving scenario, various annotators need to manually judge whether the target in the point cloud data is a moving target or a static target. The improvement is to first automatically distinguish between moving targets and static targets, and then annotate the targets. Through interpolation calculation after coordinate system transformation, the automatic judgment of moving targets and static targets is realized, greatly reducing the workload of subsequent annotators, eliminating the manual judgment process, and overall improving the overall annotation efficiency of subsequent 3D point cloud targets.

[0051] (2) The present invention can achieve interpolation prediction for the target to be annotated according to the annotation situation. Specifically, based on 3D point cloud data, interpolation means are used to automatically distinguish between moving targets and static targets, and by implementing the prediction means for future interpolation of moving targets, an interpolation prediction result is obtained. The interpolation prediction result obtained in this solution can provide a certain degree of judgment reference value for the subsequent movement direction and position of the target, effectively improving the annotation efficiency of 3D point cloud targets. Description of the Drawings

[0052] Figure 1 is a flow chart of the annotation method of the present invention. Detailed Embodiments

[0053] The present invention will be further described and explained below in conjunction with the drawings and specific embodiments. The technical features of each embodiment of the present invention can be combined correspondingly without conflict.

[0054] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings. Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below. The technical features of each embodiment of the present invention can be combined correspondingly without conflict.

[0055] Embodiment

[0056] Combined with Figure 1 shown, this embodiment provides a method for distinguishing and annotating moving and static targets of point cloud data. It should be noted that Figure 1 the solid line box belongs to the core-recorded solution of this method, and the dashed line box does not belong to the core-recorded solution of this method. The specific steps are as follows:

[0057] S1. Obtain all 3D point cloud data to be annotated;

[0058] In this embodiment, based on the visualization interface, the point cloud data of a certain target is obtained by using existing means, and any range of the target to be labeled is selected from these point cloud data. In some embodiments, the formation process of the visualization interface is as follows: according to the coordinates of any 3D point cloud data at the upper left corner and the coordinates of any 3D point cloud data at the lower right corner in the same frame within the boxed range, a two-dimensional rectangular box on the X-axis and Y-axis planes is respectively formed, and all 3D point cloud data within the range of this two-dimensional rectangular box is traversed to determine the minimum value min_z and the maximum value max_z of all 3D point cloud data in the Z-axis direction within the range of this two-dimensional rectangular box. The difference between the minimum value min_z and the maximum value max_z is used as the height of the three-dimensional rectangular box, and the activated three-dimensional rectangular box, that is, the visualization interface, is generated.

[0059] S2. Convert the 3D point cloud data in the current frame point cloud coordinate system into the 3D point cloud data in the world coordinate system; change the original coordinate values in the X-axis, Y-axis, and Z-axis directions, and convert the entire coordinate field so that subsequent judgments can be made in the same coordinate system.

[0060] In this embodiment, for the definition of the coordinate system: the current frame point cloud coordinate system refers to the coordinate system of the 3D point cloud acquisition device loaded on the vehicle itself. With the optical center of the lidar as the origin and the optical axis as the Z-axis, the position of this origin is relatively variable. The world coordinate system refers to a global reference coordinate system, which is usually fixed, and the positions and postures of all objects are described relative to this coordinate system; among them, a certain fixed position is selected as the original point of this origin, which is fixed and unchanged.

[0061] In this embodiment, the transformation matrix is composed of a rotation matrix \(R\) and a translation vector \(T\). The rotation matrix describes the rotation of the current frame point cloud coordinate system relative to the world coordinate system, and the translation vector describes the position of the origin of the current frame point cloud coordinate system in the world coordinate system.

[0062] Specifically, it can be expressed by the code snippet: \(\mathbf{T}=\begin{bmatrix}R&T\\0&1\end{bmatrix}\). Where: \(R\) is a 3×3 rotation matrix; \(T\) is a 3×1 translation vector; after \(R\) and \(T\) are combined, a 4×4 transformation matrix is formed.

[0063] In this embodiment, the specific process of transformation using the coordinate transformation formula is as follows: Assume that in the current frame point cloud coordinate system, the coordinate of a certain point is \(\mathbf{=[x,y,z,1]}^{R&T}\), and when it is transformed into the world coordinate system, \(\mathbf{'=[x',y',z',1]}^{R&T}\) is obtained.

[0064] The transformation formula is:

[0065] ;

[0066] Through the above process, the target point positions in the current frame point cloud coordinate system can be accurately transferred to the world coordinate system.

[0067] S3. Perform interpolation calculation on the converted 3D point cloud data;

[0068] In this embodiment, before the interpolation calculation, it is necessary to perform grid processing first. The specific process is as follows: Convert the 3D point cloud data facing the same target into a grid form, that is, divide the 3D point cloud data into regular grid cells. The voxel grid method is used to divide the 3D point cloud data into cubic grid cells. The specific process is as follows:

[0069] 1) Determine the voxel size: According to the range and resolution of the 3D point cloud data, determine the voxel size, which determines the grid fineness.

[0070] 2) Create a voxel grid: According to the range of the 3D point cloud data and the voxel size, create an empty voxel grid (three-dimensional array), and each element represents a voxel.

[0071] 3) Map the 3D point cloud data to the voxel grid: Traverse each point in the point cloud data, map its coordinates to the nearest voxel, and use the kd-tree nearest neighbor search algorithm to find the nearest voxel.

[0072] 4) Generate a grid from the voxel grid: Generate a grid according to the voxel state (whether there is point cloud data) in the voxel grid.

[0073] In this embodiment, based on the generated grid, bilinear interpolation is used to perform interpolation calculation, and moving targets and static targets are given according to the calculation results. The specific idea of using bilinear interpolation for interpolation calculation is as follows: First, assume that the values Z 00 , Z 10 , Z 01 , Z 11 of four known grid points Q0, Q1, Q2, Q3 are given, as well as the relative position of the target interpolation point P(tx, ty). It is required to find the value Z of the interpolation point P(tx, ty), where m = tx and n = ty. Generally, tx and ty respectively correspond to the fractional parts of m and n. Then, set the coordinates of the four grid points as: Q0: (0,0), Q1: (1,0), Q2: (0,1), Q3: (1,1); the coordinates of the interpolation point P(tx, ty) are: P(tx, ty), where tx, ty ∈ [0,1]; estimate z according to tx and ty.

[0074] Furthermore, the interpolation calculation process is as follows:

[0075] Step 31: Interpolate in the x direction to calculate P1 and P2.

[0076] Interpolate between Q0 and Q1 to calculate the value of P1(tx, 0): P1(tx, 0) = Z 00 × (1 - tx) + Z 10 × tx;

[0077] Similarly, interpolate between Q2 and Q3 to calculate the value of P2(tx, 1): P2(tx, 1) = Z 01 × (1 - tx) + Z 11 × tx;

[0078] Step 32: Interpolate in the y direction and calculate P(tx, ty) based on P1 and P2.

[0079] Interpolate between P1(tx, 0) and P2(tx, 1) to calculate the value of P(tx, ty):

[0080] P(tx, ty) = P1(tx, 0) × (1 - ty) + P2(tx, 1) × ty;

[0081] Substitute the expressions of P1(tx, 0) and P2(tx, 1) into the above formula to get:

[0082] P(tx, ty) = [Z 00 × (1 - tx) + Z 10 × tx] (1 - ty) + [Z 01 × (1 - tx) + Z 11 × tx] × ty;

[0083] Step 33: Expand and simplify:

[0084] P(tx, ty) = Z 00 × (1 - tx) × (1 - ty) + Z 10 × tx × (1 - ty) + Z 01 × (1 - tx) × ty + Z 11 × tx × ty.

[0085] Step 34: The final formula is:

[0086] P(tx, ty) = Z 00 × (1 - tx) × (1 - ty) + Z 10 × tx × (1 - ty) + Z 01 × (1 - tx) × ty + Z 11 × tx × ty.

[0087] If P(tx, ty) = 0, then the target is identified as a static target; if P(tx, ty) ≠ 0, then the target is identified as a moving target. This provides a basis for subsequent single-frame 3D point cloud data annotation of static targets and multi-frame 3D point cloud data annotation of moving targets.

[0088] S4. Based on the interpolation of various types of targets in the known 3D point cloud data, predict subsequent interpolation and annotation.

[0089] In this embodiment, the prediction idea is as follows: perform linear interpolation between two stereo bounding boxes in the point cloud data; and the linear interpolation process considers three parameters: the size, orientation angle, and center point position of the stereo bounding box. By calculating the difference values corresponding to each of these three parameters, predict the center point position, size, and orientation angle of the subsequent target, and perform automatic interpolation and annotation.

[0090] The specific prediction process is as follows:

[0091] Step S41: Data preparation; specifically, obtain the parameters of stereo bounding box A and stereo bounding box B, and these parameters include the center point position, size, and orientation angle; among them, the center point position is: A_center(x1, y1, z1), B_center(x2, y2, z2); the size is: A_size(l1, w1, h1), B_size(l2, w2, h2); the orientation angle is: A_orientation(θ1), B_orientation(θ2); where, A_center(x1, y1, z1) is the center point position of stereo bounding box A, B_center(x2, y2, z2) is the center point position of stereo bounding box B; l1 is the length of stereo bounding box A, w1 is the width of stereo bounding box A, h1 is the height of stereo bounding box A; l2 is the length of stereo bounding box B, w2 is the width of stereo bounding box B, h2 is the height of stereo bounding box B; A _ orientation(θ1) is the orientation angle of stereo bounding box A, B _ orientation(θ2) is the orientation angle of stereo bounding box B.

[0092] Step S42: Calculate the difference values of each parameter;

[0093] Calculate the difference value of the center point position:

[0094] Δx = x2 - x1;

[0095] Δy = y2 - y1;

[0096] Δz = z2 - z1;

[0097] Calculate the difference value of the size:

[0098] Δl = l2 - l1;

[0099] Δw = w2 - w1;

[0100] Δh = h2 - h1;

[0101] Calculate the difference value of the orientation angle:

[0102] Δθ = θ2 - θ1; where θ1 is the angle parameter of the three-dimensional box A, and θ2 is the angle parameter of the three-dimensional box B;

[0103] Step S43: Determine the number of interpolation steps; specifically, according to the number of interpolation targets N required, determine the number of interpolation steps: number of steps = 1 / (N + 1);

[0104] Step S44: Perform linear interpolation; specifically, for each interpolation target i (i = 1, 2,..., N), calculate the interpolation parameter, and the process is as follows:

[0105] Interpolation factor t = i × number of steps;

[0106] Interpolation of the center point position:

[0107] x i = x1 + t × Δx;

[0108] y i = y1 + t × Δy;

[0109] z i = z1 + t × Δz;

[0110] Interpolation of the size:

[0111] l i = l1 + t × Δl;

[0112] w i = w1 + t × Δw;

[0113] h i = h1 + t × Δh;

[0114] Interpolation of the orientation angle:

[0115] θ i = θ1 + t × Δθ;

[0116] Step S45: Generate interpolation targets; specifically, according to the parameters calculated by interpolation, the three-dimensional box parameters of the interpolation targets are:

[0117] Center point position of interpolation target i: (x_i, y_i, z_i);

[0118] Size of interpolation target i: (l_i, w_i, h_i);

[0119] Orientation angle of the interpolation target i: θ_i;

[0120] Step S46: Output the interpolation result; specifically, output the parameters of all interpolation target bounding boxes to form an interpolation result set.

[0121] In this embodiment, for static targets, the known data interpolation is 0, 0, 0…, and then the subsequent predicted interpolation is still 0, 0, 0….

[0122] In this embodiment, for moving targets, assuming that the target rotates and translates at a constant speed, based on the known interpolation of the target, the interpolation of the target within a certain period of time can be predicted. If the known target interpolation is 1, 2, 3…, then the predicted interpolation can be 4, 5, 6…. It should be noted that only constant speed, that is, the analysis of static and moving targets in the case of uniform speed, is considered in this embodiment.

[0123] Embodiment 2

[0124] Based on the annotation method of Embodiment 1, this embodiment provides a static and moving target discrimination annotation system for point cloud data, including an annotation module, a coordinate system conversion module, an interpolation calculation module, and an interpolation prediction module, where,

[0125] The annotation module is a tool for starting the annotation of 3D point cloud data, determining the coordinate values in the x-axis, y-axis, and z-axis directions respectively, starting the three-dimensional rectangular box annotation, and loading each 3D point cloud data to be annotated based on the visualization display interface, facilitating subsequent data processing and other operations.

[0126] The coordinate system conversion module is mainly used to convert the current frame point cloud coordinate system originally defaulted by the system into the world coordinate system, change the original coordinate values in the x-axis, y-axis, and z-axis directions, and convert the entire coordinate field of view.

[0127] Based on the world coordinate system, the interpolation calculation module performs interpolation calculation on the 3D point cloud data of the same target in different frames, uses the linear interpolation method to calculate the relative distance between each frame of the target (dx, dy, dz) and the origin (x(0), y(0), z(0)). According to the read interpolation calculation result, it is judged whether the target is a static target or a moving target. If the target is a static target, the change in the target position on different frames is close to 0, so the interpolation in the time series can be expressed as "0, 0, 0..."; if the target is a moving target, the change in the target position on different frames is obvious and shows regular changes, so the interpolation in the time series can be expressed as "1, 2, 3..."; and then it is judged whether the target is a moving target or a static target, providing a reference for the subsequent annotation work. Further, for static targets, for a single frame of 3D point cloud data, single-frame point cloud data annotation can be used to carry out the annotation work; for moving targets, for 2-3 frames of 3D point cloud data, multi-frame point cloud data annotation can be used to carry out the annotation work.

[0128] The interpolation prediction module is mainly aimed at moving targets. According to the time series interpolation rule between known data, it predicts the interpolation of the target in the subsequent time series, facilitating the subsequent annotation actions.

[0129] In some other embodiments, a system for distinguishing and annotating static and dynamic targets of point cloud data is also provided. This system is used to execute the specific process of the method for distinguishing and annotating static and dynamic targets of point cloud data.

[0130] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than limiting the protection scope of the present invention. Any simple modification or equivalent replacement of the technical solution of the present invention by those of ordinary skill in the art does not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A method for distinguishing and labeling static and dynamic targets in point cloud data, characterized in that It includes the following steps: S1. Obtain all 3D point cloud data to be labeled; Among them, in S1, based on the visualization interface, obtain the 3D point cloud data of a certain target, and select any range of the target to be labeled from these 3D point cloud data; The formation process of the visualization interface is as follows: According to the coordinates of any 3D point cloud data at the upper left corner and the coordinates of any 3D point cloud data at the lower right corner in the same frame within the boxed range, form two-dimensional rectangular frames on the X-axis and Y-axis planes respectively, and traverse all 3D point cloud data within the range of this two-dimensional rectangular frame to determine the minimum value min_z and the maximum value max_z of all 3D point cloud data within the range of this two-dimensional rectangular frame in the Z-axis direction. Take the difference between the minimum value min_z and the maximum value max_z as the height of the three-dimensional rectangular frame, and generate the activated three-dimensional rectangular frame, which is the visualization interface; S2. Convert the 3D point cloud data in the current frame point cloud coordinate system into 3D point cloud data in the world coordinate system; S3. Perform interpolation calculation on the converted 3D point cloud data; Among them, in the interpolation calculation process, set the target interpolation point P(tx, ty). If P(tx, ty)=0, then determine that this target is a static target; if P(tx, ty)≠0, then determine that this target is a moving target; S4. Based on the interpolation of various types of targets in the 3D point cloud data, predict subsequent interpolation and labeling; Among them, the prediction method is: perform linear interpolation between two stereo frames in the point cloud data; and the linear interpolation process considers three parameters: the size, orientation angle, and center point position of the stereo frame. By calculating the difference values corresponding to these three parameters respectively, predict the center point position, size, and orientation angle of the subsequent target, and perform automatic interpolation and labeling.

2. A method for distinguishing and labeling moving and static targets of point cloud data according to claim 1, characterized in that, In S4, the prediction process is as follows: Step S41: Data preparation; Step S42: Calculate the difference values of each parameter; Step S43: Determine the interpolation steps; Step S44: Perform linear interpolation; Step S45: Generate interpolation targets; Step S46: Output the interpolation results.

3. A method for distinguishing and labeling moving and static targets of point cloud data according to claim 2, characterized in that, In step S41, the data are the parameters of stereo frame A and stereo frame B, and these parameters include the center point position, size, and orientation angle; The center point positions are: A_center(x1,y1,z1), B_center(x2,y2,z2); The size is: A _ size(l1, w1, h1), B _ size(l2, w2, h2); The orientation angles are: A_orientation(θ1), B_orientation(θ2); Among them, A_center(x1, y1, z1) is the center point position of the three-dimensional box A, and B_center(x2, y2, z2) is the center point position of the three-dimensional box B; l1 is the length of the three-dimensional box A, w1 is the width of the three-dimensional box A, and h1 is the height of the three-dimensional box A; l2 is the length of the three-dimensional box B, w2 is the width of the three-dimensional box B, and h2 is the height of the three-dimensional box B; A _ orientation(θ1) is the orientation angle of the three-dimensional box A, and B_orientation(θ2) is the orientation angle of the three-dimensional box B.

4. A method for distinguishing and labeling static and dynamic targets of point cloud data according to claim 3, characterized in that, In step S42, the calculation process of the difference values is as follows: Calculate the difference value of the center point position: Δx = x2 - x1; Δy = y2 - y1; Δz = z2 - z1; Calculate the difference value of the size: Δl = l2 - l1; Δw = w2 - w1; Δh = h2 - h1; Calculate the difference value of the orientation angle: Δθ = θ2 - θ1; where θ1 is the angle parameter of stereo frame A and θ2 is the angle parameter of stereo frame B.

5. A method for distinguishing and labeling moving and static targets of point cloud data according to claim 4, characterized in that, In step S43, according to the number N of targets to be interpolated, determine the interpolation steps: steps = 1 / (N + 1).

6. A method for distinguishing and labeling static and dynamic targets of point cloud data according to claim 5, characterized in that, In step S44, for each interpolation target i, interpolation parameters are calculated, where i = 1, 2,..., N; The process is as follows: Interpolation factor t = i × number of steps; Interpolation of the center point position: x i = x1 + t × Δx; y i = y1 + t × Δy; z i = z1 + t × Δz; Interpolation of the size: l i = l1 + t × Δl; w i = w1 + t × Δw; h i = h1 + t × Δh; Orientation angle interpolation: θ i = θ1 + t × Δθ.

7. A method for distinguishing and labeling static and dynamic targets of point cloud data according to claim 6, characterized in that, In step S45, according to the parameters calculated by interpolation, the stereo box parameters of the interpolation target are: Center point position of interpolation target i: (x_i, y_i, z_i); Size of interpolation target i: (l_i, w_i, h_i); Orientation angle of interpolation target i: θ_i.

8. A static and dynamic target discrimination and annotation system for point cloud data, characterized in that, Used to execute a method for distinguishing and labeling static and dynamic targets in point cloud data according to any one of claims 1-7; Including a labeling module, a coordinate system conversion module, an interpolation calculation module, and an interpolation prediction module; The labeling module is used to start the 3D point cloud data labeling; The coordinate system conversion module is used to convert the current frame point cloud coordinate system into the world coordinate system; The interpolation calculation module performs interpolation calculation on the 3D point cloud data of the same target in different frames loaded based on the world coordinate system; The interpolation prediction module predicts the interpolation of the moving target in the subsequent time series for the moving target and performs the labeling action.

Images