A boundary line filtering method applied to a mine unstructured road scene

Abstract

The application discloses a boundary line filtering method applied to a mine unstructured road scene, and comprises the following steps: firstly, performing soft synchronization on solid-state laser radar transmission data and PCL point cloud library screen point tool selected road boundary point clouds; then, performing cubic spline interpolation preprocessing on obtained boundary line data information, so that the point cloud densities of two road boundary lines are roughly the same; performing time stamp synchronous receiving on the laser radar point cloud data information and the above-mentioned preprocessed boundary line data information; finally, performing filtering processing on the received boundary line data information and the current frame of laser radar point cloud information, obtaining road point cloud information between the two boundary lines after the filtering processing; in the unstructured road scene, the three-dimensional point clouds required by a curved road surface can be effectively extracted, the extraction efficiency is high, useless point cloud data and redundant data can be removed, and the bandwidth pressure during subsequent point cloud data transmission is effectively reduced.

Description

Technical Field

[0001] This invention relates to a boundary line filtering method, specifically a boundary line filtering method applied to unstructured road scenarios in mines. Background Technology

[0002] LiDAR is now widely used in autonomous driving and geographic mapping. Its main advantage is that it can generate 3D location information to quickly determine the location, size, appearance, and material of objects, while acquiring data to form accurate digital models. Compared with sensors such as cameras, it has a longer detection range, higher accuracy, faster response speed, and is unaffected by ambient light. However, the 3D point cloud information generated by LiDAR often contains hundreds of thousands or even millions of 3D points. Not all of these 3D point information is useful. Currently, autonomous driving mainly needs to acquire points on the road surface, and the large number of 3D points generated increases bandwidth pressure when transmitting this information. Therefore, to remove useless points from the generated 3D point cloud and retain the required points, filtering is performed. Currently, the most widely used filtering methods are pass-through filtering and voxel filtering. Pass-through filtering is suitable for scenes parallel to the coordinate axes, but it is less effective for curved scenes in autonomous driving, often losing a large amount of the required point cloud for the road surface area at curves. Voxel filtering cannot remove redundant scenes; although it can reduce the number of 3D points, its efficiency in extracting the required point cloud for the road surface area is still relatively poor. Compared to structured roads, unstructured roads in mines are characterized by large curves and irregular boundary lines; therefore, both of the above filtering methods are less effective in unstructured road scenarios.

[0003] Therefore, proposing a new filtering method that can effectively extract the required 3D point cloud of curved road surfaces in unstructured road scenarios with high extraction efficiency, and can remove useless and redundant point cloud data, thereby effectively reducing the bandwidth pressure when transmitting point cloud data, is one of the research directions in this industry. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a boundary line filtering method applicable to unstructured road scenarios in mines. In unstructured road scenarios, it can effectively extract the 3D point cloud required for curved road surfaces with high extraction efficiency, and can remove useless point cloud data and redundant data, thereby effectively reducing the bandwidth pressure when transmitting point cloud data subsequently.

[0005] To achieve the above objectives, the technical solution adopted by this invention is: a boundary line filtering method applied to unstructured road scenarios in mines. A sensing system is installed on a mining truck. The sensing hardware system includes a solid-state lidar and a computer. The computer includes a PCL point cloud library screen selection tool, a timestamp synchronization module, a boundary line point cloud preprocessing module, a filtering module, and a display. The solid-state lidar transmits the sensed 3D point cloud data to the computer. The PCL point cloud library screen selection tool selects road boundary point clouds from the 3D point cloud data to obtain boundary line data. The computer then performs cubic spline interpolation preprocessing on the acquired boundary line data through the boundary line point cloud preprocessing module, ensuring that the point cloud densities of the two boundary lines are approximately the same after preprocessing. The timestamp synchronization module synchronizes the timestamps of the solid-state lidar point cloud with the preprocessed boundary lines. The filtering module uses the boundary line information to filter the lidar point cloud of the current frame to obtain road point cloud information between the two boundary lines. Finally, the obtained road point cloud information between the two boundary lines is transmitted to the display for display. The specific filtering steps are as follows:

[0006] A. Perform soft synchronization between the solid-state lidar transmission data and the road boundary point cloud selected by the PCL point cloud library screen point selection tool to ensure that the lidar point cloud data information and boundary line data information are released at the same time.

[0007] B. Perform cubic spline interpolation preprocessing on the boundary line data information obtained in step A to make the point cloud density of the two road boundary lines approximately the same.

[0008] C. The computer timestamps and receives the lidar point cloud data information obtained in step A and the boundary line data information after preprocessing in step B.

[0009] D. The received boundary line data information and the lidar point cloud information of the current frame are filtered. The filtering process is as follows: obtaining the bounding box, point cloud translation, grid projection, ray projection, and superposition of point cloud information within the grid. After the filtering process is completed, the road point cloud information between the two boundary lines is obtained.

[0010] Furthermore, the soft synchronization method in step A is as follows: In the ROS system, before the solid-state LiDAR and the PCL point cloud library screen selection tool publish the data, they first stamp the timestamp with a unified Greenwich Mean Time.

[0011] Furthermore, the specific process of performing cubic spline interpolation preprocessing on the boundary line data information in step B is as follows:

[0012] First, the boundary line data [a,b] is divided into n intervals, namely [(x 0, x1),(x1, x2),(x 2, x3),…,(x n-1 ,x n There are a total of n+1 points, with two endpoints x0 = a and x1 = a. n =b; A cubic spline is a curve for each interval that is a cubic equation. The cubic spline equation satisfies the following conditions:

[0013] (1) In each interval (x) i, x i+1 S(x) = S i (x) are all cubic equations;

[0014] (2) The interpolation condition is satisfied, i.e., S(x) i )=y i , i = 0, 1, 2, 3, ..., n;

[0015] (3) The curve is smooth, that is, S(x), S′(x), S″(x) are continuous;

[0016] Given the above conditions, the cubic equation for each interval can be constructed in the form y = a i +b i x+c i x 2 +d i x 3 This form is the cubic spline function S. i (x); the unknown coefficients (a) for each interval i ,b i ,c i ,d i If we solve this problem, there are 4n unknowns across n intervals, requiring 4n equations to solve. By obtaining these 4n equations, we can find the four unknown coefficients for each interval, and thus derive the cubic spline function S for each interval. i (x), and finally the cubic spline function S for each interval. i (x) Interpolate each interval according to the set step size.

[0017] Furthermore, the specific process for obtaining the 4n equations is as follows:

[0018] Based on the interpolation conditions in (2) above, except for the two endpoints, all n-1 internal points satisfy S i (x i+1 )=y i+1 ,S i+1 (x i+1 )=y i+1That is, the cubic equations in the two adjacent intervals of each interior point result in 2(n-1) equations. Adding the cubic equations of the two endpoints satisfying the first and last intervals respectively, there are a total of 2n equations. Secondly, according to the condition of curve smoothness in (3), the first derivatives of the n-1 interior points are continuous, meaning the end point of the i-th interval and the starting point of the i+1-th interval are the same point. Therefore, the first derivatives at the same point are equal, i.e., S... i ′(x i+1 ) = S i+1 ′(x i+1 If ), then there are n-1 equations; furthermore, the second derivative at interior points is also continuous, i.e., S i "(x i+1 ) = S i+1 "(x i+1 There are also n-1 equations, so there are a total of 4n-2 equations. The remaining two equations are obtained through boundary conditions:

[0019] Natural Spline: The second derivative at the specified endpoints is 0, i.e., S″(x0)=0=S″(x0) n );

[0020] Clamped Spline: Specifies the first derivatives at the endpoints, here defined as A and B respectively, i.e., S0′(x0)=A,S n-1 ′(x n ) = B;

[0021] Non-Knot Spline: The third derivative of the first interpolation point is forced to be equal to the third derivative of the second point, and the third derivative of the last point is forced to be equal to the third derivative of the second-to-last point, i.e., S0″′(x0)=S1″′(x1), S n-2 "′(x n-1 ) = S n-1 "′(x n Thus, 4n equations are obtained.

[0022] Furthermore, the specific process of filtering in step D is as follows:

[0023] ① Calculate the bounding box: Sort the points in the current frame's LiDAR point cloud according to the x, y, z directions. Find the maximum and minimum values ​​in the x-direction, y-direction, and z-direction. Calculate the dimensions of the bounding box as follows:

[0024]

[0025] In the above formula, max x This represents the maximum value in the x-direction, min.x This represents the minimum value in the x-direction; max y This represents the maximum value in the y-direction, min. y This represents the minimum value in the y-direction; max z This represents the maximum value in the z-direction, min. z This represents the minimum value in the z-direction; abs indicates taking the absolute value.

[0026] ② Point cloud translation: To ensure that the point cloud index is positive after being projected onto the raster, the point cloud needs to be translated based on the minimum values ​​of x and y, so that the coordinates of the points in the point cloud do not have negative values. The specific formula is as follows:

[0027]

[0028] ③ Raster Projection: Using the bounding box size obtained in step ① and the set resolution parameter, calculate the raster index and sequence number corresponding to each point in the point cloud. The formula for calculating the raster index is as follows:

[0029]

[0030] Cell x With Cell y The x and y values ​​represent the raster indices, and Ceil represents the rounding function. The raster index for each 3D point is calculated using this function. After calculating the raster index, to ensure it remains within the projection range, raster cells outside the range are deleted. First, the calculated Size is used... x Size y The map size of the raster is calculated by combining the resolutions:

[0031]

[0032] The selection criteria are as follows:

[0033]

[0034]

[0035] Filter the rasters using the above method, retaining those that meet the criteria, and then calculate the raster sequence number based on the calculated raster index:

[0036] Cell num =Cell x *Len y +Cell y

[0037] Perform raster projection as described above, that is, project the 3D point cloud onto the 2D raster and obtain the raster index and raster number corresponding to each 3D point;

[0038] ④ Ray casting: After projecting the two boundary lines onto the two-dimensional grid map, obtain the grid between the left and right boundary lines. Then, perform ray casting on each of the grids projected onto the left and right boundary lines. After completion, obtain the grid index between the left and right boundary lines.

[0039] ⑤ Overlay of point cloud information within the raster: Using the raster index between the two boundary lines obtained in step ④, the projection points obtained in step ③ within the corresponding raster are overlaid to obtain the overlaid point cloud. Since coordinate translation was performed in step ②, in order to maintain the original coordinates, the coordinates are transformed back using the following formula:

[0040]

[0041] After coordinate transformation, the point cloud results are saved, which yields the filtered road point cloud data.

[0042] Furthermore, the specific steps for light projection in step ④ are as follows:

[0043] (1) Give the two raster indices (Left) of the left and right boundary lines. x Left y ),(Right x Right y Place the point on the left into the point (x0, y0);

[0044] (2) Load (x0, y0) into the cache container to obtain the first point;

[0045] (3) Calculate the constants Δx, Δy, 2Δy, and 2Δy-2Δx, where Δx and Δy are the absolute values ​​of the differences between the x and y coordinates of the first and last points, respectively. Calculate the first value of the decision parameter. The formula for the decision parameter is as follows:

[0046] p0 = 2Δy - Δx

[0047] (4) Starting from k=0, at each x along the path k The following tests will be performed at this location:

[0048] If p k If x < 0, then the next point to be drawn is (x k +1,y k ),and

[0049] p k+1 =p k +2Δy

[0050] Otherwise, the next point to be drawn is (x k +1,y k +1), and

[0051] p k+1 =p k +2Δy-2Δx

[0052] (5) Repeat step (4) a total of Δx-1 times to obtain the grid index between the left and right boundary lines.

[0053] Compared with existing technologies, this invention first performs soft synchronization between the solid-state lidar transmission data and the road boundary point cloud selected by the PCL point cloud library screen point selection tool, ensuring that the lidar point cloud data and boundary line data are released at the same time. Then, the acquired boundary line data is preprocessed with cubic spline interpolation to make the point cloud density of the two road boundary lines approximately the same. The acquired lidar point cloud data and the preprocessed boundary line data are received synchronously with timestamps. Finally, the received boundary line data and the lidar point cloud information of the current frame are filtered to obtain the road point cloud information between the two boundary lines. In unstructured road scenarios, the extracted road boundary lines are used to filter the acquired point cloud image, thereby filtering out the road point cloud data between the two boundary lines, obtaining the required road point cloud data, and removing the remaining useless and redundant point cloud data, reducing redundant points, improving the processing speed of downstream algorithms, and avoiding the use of other redundancy removal algorithms. This makes the entire processing process highly efficient and meets the required data processing requirements, while effectively reducing the bandwidth pressure when transmitting point cloud data later. Attached Figure Description

[0054] Figure 1 This is an overall flowchart of the present invention;

[0055] Figure 2 This is a diagram illustrating the filtering effect of the present invention on a U-shaped bend;

[0056] Figure 3 This is a diagram showing the filtering effect of the present invention on a straight path. Detailed Implementation

[0057] The present invention will be further described below.

[0058] like Figure 1As shown, this invention first equips a sensing system on a mining truck. The sensing hardware system includes a solid-state lidar and a computer. The computer includes a PCL point cloud library screen selection tool, a timestamp synchronization module, a boundary line point cloud preprocessing module, a filtering module, and a display. The solid-state lidar is used to transmit the sensed 3D point cloud data to the computer. The PCL point cloud library screen selection tool is used by the operator to manually select road boundary point clouds from the 3D point cloud data to obtain boundary line data. Then, the computer performs cubic spline interpolation preprocessing on the acquired boundary line data through the boundary line point cloud preprocessing module, so that the point cloud densities of the two boundary lines are approximately the same after preprocessing. The timestamp synchronization module is used to receive the point cloud from the solid-state lidar and the preprocessed boundary lines synchronously with timestamps. The filtering module is used to filter the lidar point cloud of the current frame using the boundary line information to obtain road point cloud information between the two boundary lines. Finally, the acquired road point cloud information between the two boundary lines is transmitted to the display for display. The specific filtering steps are as follows:

[0059] A. Perform soft synchronization between the solid-state lidar transmission data and the road boundary point cloud selected by the PCL point cloud library screen selection tool to ensure that the lidar point cloud data information and boundary line data information are released at the same time. The soft synchronization method is as follows: in the ROS system, before the solid-state lidar and the PCL point cloud library screen selection tool release their data, they are stamped with a unified Greenwich Mean Time.

[0060] B. Perform cubic spline interpolation preprocessing on the boundary line data obtained in step A to ensure that the point cloud density of the two road boundary lines is approximately the same. The specific process of cubic spline interpolation preprocessing is as follows:

[0061] First, the boundary line data [a,b] is divided into n intervals, namely [(x 0, x1),(x 1, x2),(x 2, x3),…,(x n-1 ,x n There are a total of n+1 points, with two endpoints x0 = a and x1 = a. n =b; A cubic spline is a curve for each interval that is a cubic equation. The cubic spline equation satisfies the following conditions:

[0062] (1) In each interval (x) i, x i+1 S(x) = S i (x) are all cubic equations;

[0063] (2) The interpolation condition is satisfied, i.e., S(x) i )=y i, i = 0, 1, 2, 3, ..., n;

[0064] (3) The curve is smooth, that is, S(x), S′(x), S″(x) are continuous;

[0065] Given the above conditions, the cubic equation for each interval can be constructed in the form y = a i +b i x+c i x 2 +d i x 3 This form is the cubic spline function S. i (x); the unknown coefficients (a) for each interval i ,b i ,c i ,d i If we solve this problem, then there are 4n unknowns in the n intervals, which means we need 4n equations to solve it. The specific process of obtaining these 4n equations is as follows:

[0066] Based on the interpolation conditions in (2) above, except for the two endpoints, all n-1 internal points satisfy S i (x i+1 )=y i+1 ,S i+1 (x i+1 )=y i+1 That is, the cubic equations in the two adjacent intervals of each interior point result in 2(n-1) equations. Adding the cubic equations of the two endpoints satisfying the first and last intervals respectively, there are a total of 2n equations. Secondly, according to the condition of curve smoothness in (3), the first derivatives of the n-1 interior points are continuous, meaning the end point of the i-th interval and the starting point of the i+1-th interval are the same point. Therefore, the first derivatives at the same point are equal, i.e., S... i ′(x i+1 ) = S i+1 ′(x i+1 If ), then there are n-1 equations; furthermore, the second derivative at interior points is also continuous, i.e., S i "(x i+1 ) = S i+1 "(x i+1 There are also n-1 equations, so there are a total of 4n-2 equations. The remaining two equations are obtained through boundary conditions:

[0067] Natural Spline: The second derivative at the specified endpoints is 0, i.e., S″(x0)=0=S″(x0) n );

[0068] Clamped Spline: Specifies the first derivatives at the endpoints, here defined as A and B respectively, i.e., S0′(x0)=A,S n-1 ′(x n ) = B;

[0069] Non-Knot Spline: The third derivative of the first interpolation point is forced to be equal to the third derivative of the second point, and the third derivative of the last point is forced to be equal to the third derivative of the second-to-last point, i.e., S0″′(x0)=S1″′(x1), S n-2 "′(x n-1 ) = S n-1 "′(x n This yields 4n equations. Then, based on these 4n equations, the four unknown coefficients for each interval are determined, leading to the cubic spline function S for each interval. i (x), and finally the cubic spline function S for each interval. i (x) Interpolate each interval according to the set step size.

[0070] C. The computer timestamps and receives the lidar point cloud data information obtained in step A and the boundary line data information after preprocessing in step B.

[0071] D. Filter the received boundary line data information and the LiDAR point cloud information of the current frame. The specific filtering process is as follows: ① Calculate the bounding box: Sort the points in the LiDAR point cloud of the current frame according to the order of x, y, z, and find the maximum and minimum values ​​in the x-direction, y-direction, and z-direction. Calculate the size of the bounding box as follows:

[0072]

[0073] In the above formula, max x This represents the maximum value in the x-direction, min. x This represents the minimum value in the x-direction; max y This represents the maximum value in the y-direction, min. y This represents the minimum value in the y-direction; max z This represents the maximum value in the z-direction, min. z This represents the minimum value in the z-direction; abs indicates taking the absolute value.

[0074] ② Point cloud translation: To ensure that the point cloud index is positive after being projected onto the raster, the point cloud needs to be translated based on the minimum values ​​of x and y, so that the coordinates of the points in the point cloud do not have negative values. The specific formula is as follows:

[0075]

[0076] ③ Raster Projection: Using the bounding box size obtained in step ① and the set resolution parameter, calculate the raster index and sequence number corresponding to each point in the point cloud. The formula for calculating the raster index is as follows:

[0077]

[0078] Cell x With Cell y The x and y values ​​represent the raster indices, and Ceil represents the rounding function. The raster index for each 3D point is calculated using this function. After calculating the raster index, to ensure it remains within the projection range, raster cells outside the range are deleted. First, the calculated Size is used... x Size y The map size of the raster is calculated based on the resolution:

[0079]

[0080] The selection criteria are as follows:

[0081]

[0082]

[0083] Filter the rasters using the above method, retaining those that meet the criteria, and then calculate the raster sequence number based on the calculated raster index:

[0084] Cell num =Cell x *Len y +Cell y

[0085] Perform raster projection as described above, that is, project the 3D point cloud onto the 2D raster and obtain the raster index and raster number corresponding to each 3D point;

[0086] ④ Ray casting: After projecting the two boundary lines onto the 2D grid map, obtain the grid between the left and right boundary lines. Then, perform ray casting on each of the grids projected onto the left and right boundary lines. The specific steps of ray casting are as follows:

[0087] (1) Give the two raster indices (Left) of the left and right boundary lines. x Left y ),(Right x Right y Place the point on the left into the point (x0, y0);

[0088] (2) Load (x0, y0) into the cache container to obtain the first point;

[0089] (3) Calculate the constants Δx, Δy, 2Δy, and 2Δy-2Δx, where Δx and Δy are the absolute values ​​of the differences between the x and y coordinates of the first and last points, respectively. Calculate the first value of the decision parameter. The formula for the decision parameter is as follows:

[0090] p0 = 2Δy - Δx

[0091] (4) Starting from k=0, at each x along the path k The following tests will be performed at this location:

[0092] If p k If x < 0, then the next point to be drawn is (x k +1,y k ),and

[0093] p k+1 =p k +2Δy

[0094] Otherwise, the next point to be drawn is (x k +1,y k +1), and

[0095] p k+1 =p k +2Δy-2Δx

[0096] (5) Repeat step (4) a total of Δx-1 times to obtain the grid index between the left and right boundary lines.

[0097] ⑤ Overlay of point cloud information within the raster: Using the raster index between the two boundary lines obtained in step ④, the projection points obtained in step ③ within the corresponding raster are overlaid to obtain the overlaid point cloud. Since coordinate translation was performed in step ②, in order to maintain the original coordinates, the coordinates are transformed back using the following formula:

[0098]

[0099] After coordinate transformation, save the point cloud results, such as... Figure 2 and 3 As shown, the filtered road point cloud data is obtained.

[0100] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A boundary line filtering method applied to a mine unstructured road scene, characterized in that, A sensing system is assembled on a mine truck, which comprises a solid-state laser radar and a computer, the computer comprising a PCL point cloud library screen point selection tool, a timestamp synchronization module, a boundary line point cloud preprocessing module, a filtering processing module and a display, the solid-state laser radar being used to transmit three-dimensional point cloud data sensed to the computer; the PCL point cloud library screen point selection tool is used for a staff to manually select road boundary point cloud from the three-dimensional point cloud data to obtain boundary line data; then the computer pre-processes the obtained boundary line data through the boundary line point cloud preprocessing module by cubic spline interpolation, so that the densities of the two pre-processed boundary line point clouds are approximately the same; the timestamp synchronization module is used to synchronously receive the point cloud of the solid-state laser radar and the pre-processed boundary line; the filtering processing module is used to filter the laser radar point cloud of the current frame by using the boundary line information to obtain road point cloud information between the two boundary lines; finally, the obtained road point cloud information between the two boundary lines is transmitted to the display for display, and the specific filtering steps are as follows: A, soft synchronization is performed on the solid-state laser radar transmission data and the road boundary point cloud selected by the PCL point cloud library screen point selection tool, specifically: before the solid-state laser radar and the PCL point cloud library screen point selection tool respectively publish data in the ROS system, a unified Greenwich time is stamped on the timestamp, so as to ensure that the laser radar point cloud data information and the boundary line data information are published at the same time respectively; B, cubic spline interpolation preprocessing is performed on the boundary line data information obtained in step A, so that the point cloud densities of the two boundary lines of the road are approximately the same; C, the computer synchronously receives the laser radar point cloud data information obtained in step A and the boundary line data information pre-processed in step B through the timestamp; D, the received boundary line data information and the laser radar point cloud information of the current frame are filtered, and the filtering process is as follows: bounding box, point cloud translation, grid projection, ray projection, and grid point cloud information superposition, so as to obtain road point cloud information between the two boundary lines after filtering.

2. The boundary line filtering method applied to the mine unstructured road scene according to claim 1, characterized in that, The specific process of the cubic spline interpolation preprocessing of the boundary line data information in step B is as follows: First, the boundary line data [a,b] is divided into n intervals, namely [(x 0, x1),(x 1, x2),(x 2, x3),…,(x n-1, x n There are a total of n+1 points, with two endpoints x0 = a and x1 = a. n =b; A cubic spline is a curve for each interval that is a cubic equation. The cubic spline equation satisfies the following conditions: (1) S(x) = S(x) on each interval (x i, x i+1 ) on which S(x) = S i (x) is a cubic equation; (2) meet the interpolation condition, i.e. S(x i ) = y i , i = 0, 1, 2, 3, …, n; (3) the curve is smooth, i.e. S'(x), S"(x) are continuous; ′ (x),S"(x) are continuous; Through the above conditions, the cubic equation of each interval can be constructed in the form y = a i +b i x+c i x 2 +d i x 3 This form is the cubic spline function S i (x); for each interval, the unknown coefficients (a i ,b i ,c i ,d i ) are solved, then there are 4n unknowns for n intervals, then 4n equations are needed to solve, and then the four unknown coefficients of each interval are obtained, and then the cubic spline function S i (x) of each interval is obtained, and finally the cubic spline function S i (x) of each interval is interpolated according to the set step.

3. The boundary line filtering method applied to the mine unstructured road scene according to claim 2, characterized in that, The specific process of obtaining the 4n equations is as follows: By the interpolation condition in (2) above, except for the two endpoints, all the n-1 interior points satisfy S i (x i+1 ) = y i+1 , S i+1 (x i+1 ) = y i+1 , i.e. each interior point satisfies a cubic equation in the two adjacent intervals, then there are 2(n-1) equations, plus the two endpoints which satisfy a cubic equation in the first and last intervals respectively, so there are a total of 2n equations; secondly, according to the smoothness condition in (3), the first derivative of the n-1 interior points is continuous, i.e. the end point of the ith interval and the start point of the ith+1 interval are the same point, so the first derivative of the same point is equal, i.e. S i ′ (x i+1 ) = S i+1 ′ (x i+1 ), so there are n-1 equations; in addition, the second derivative of the interior points is also continuous, i.e. S i ″(x i+1 ) = S i+1 ″(x i+1 ), also n-1 equations, so there are a total of 4n-2 equations, and the remaining two equations are obtained by one of the following boundary conditions: Natural bounds: specify the second derivative of the endpoints to be zero, i.e. S"(x0) = 0 = S"(x n ); Fixed boundary: specify end-point first derivative, here A and B, respectively, i.e. S0 ′ (x0) = A, S n-1 ′ (x n ) = B; Non-knotted boundary: The third derivative of the first interpolation point is forced to be equal to the third derivative of the second point, and the third derivative of the last point is forced to be equal to the third derivative of the second-to-last point, i.e., S0″′(x0)=S1 ″′ (x1), S n-2 "′(x n-1 ) = S n-1 "′(x n Thus, 4n equations are obtained.

4. The boundary line filtering method for use in a mine unstructured road scene according to claim 1, characterized in that, The specific process of the filtering in step D is as follows: ① Bounding box: the points in the current frame laser radar point cloud are sorted in the order of x, y and z, and the maximum and minimum values in the x direction, the maximum and minimum values in the y direction, and the maximum and minimum values in the z direction are calculated, and the size of the bounding box is as follows: max in x direction, min x represents the maximum value in the x direction, min x represents the minimum value in the x direction; max y represents the maximum value in the y direction, min y represents the minimum value in the y direction; max z represents the maximum value in the z direction, min z represents the minimum value in the z direction; abs represents taking the absolute value; ② Point cloud translation: in order to ensure that the point cloud index is positive after the point cloud is projected onto the grid, the point cloud is translated according to the minimum values of x and y, so that the coordinates of the points in the point cloud do not have negative values, and the specific formula is as follows: ③ Grid projection: the size of the bounding box obtained in step ① and the resolution parameter resolution are used to calculate the grid index and serial number corresponding to each point in the point cloud, and the formula for calculating the grid index is as follows: Ceil x Ceil y Ceil x Ceil y Ceil The screening criteria are as follows: According to the above manner, the grid is screened, and the grid meeting the condition is reserved, and then the grid serial number is calculated according to the calculated grid index: Cell num = Cell x * Len y + Cell y According to the above grid projection, the three-dimensional point cloud is projected onto the two-dimensional grid, and the grid index corresponding to each three-dimensional point and the grid serial number are obtained; (4) Light projection: after projecting the two boundary lines onto the two-dimensional grid map, the grid between the left and right boundary lines is obtained, and then the light projection is performed on the grid projected by the left and right boundary lines one by one, and the grid index between the left and right boundary lines is obtained after completion; (5) Superimpose the point cloud information in the grid: use the grid index between the two boundary lines obtained in step 4 to superimpose the projected points obtained in step 3 in the corresponding grid, so as to obtain the superimposed point cloud. Since the coordinate translation is performed in step 2, in order to maintain the original coordinates unchanged, the coordinates are transformed back through the following formula: After coordinate transformation, save the point cloud result, that is, obtain the filtered road point cloud data.

5. The boundary line filtering method for use in a mine unstructured road scene according to claim 4, characterized in that, The specific steps of the light projection in step 4 are as follows: (1) Give the two raster indices (Left) of the left and right boundary lines. x Left y ),(Right x Right y Place the point on the left into the point (x0, y0); (2) Put (x0, y0) into the buffer container to obtain the first point; (3) Calculate the constants Δx, Δy, 2Δy, 2Δy-2Δx, wherein Δx and Δy are the absolute values of the horizontal and vertical coordinate differences of the first and last points, and calculate the first value of the decision parameter, and the formula of the decision parameter is as follows: p0=2Δy-Δx (4) Starting from k = 0, at every x k along the path, the following detection is made: If p k <0, then the next point to draw is (x k +1,y k ), and p k+1 = p k + 2Δy Otherwise, the next point to draw is (x k +1,y k +1), and p k+1 = p k + 2Δy - 2Δx (5) Repeat step (4) for Δx-1 times, so as to obtain the grid index between the left and right boundary lines.

Images