Method and system for fusing voxel-based point cloud high-definition map fusion deviation elimination

Abstract

The application provides a voxel-based point cloud high-precision map fusion deviation elimination method and system, comprising the following steps: collecting point cloud data of a to-be-constructed map A to obtain a point cloud data set A; making a map according to the pre-fusion pose of the point cloud data set A to generate a map A; collecting point cloud data of a to-be-constructed map B to obtain a point cloud data set B; making a map according to the post-fusion pose of the point cloud data set B to generate a map B; generating a voxel description file according to the voxel coordinates; adjusting the map A according to the adjusted point cloud data set A and the post-fusion pose to obtain a map A'; and superimposing and fusing the map A' and the map B to obtain a map C. According to the pre-fusion and post-fusion poses of the mapping data, the file capable of adjusting the high-precision map is generated, the file is used, the purpose of automatically adjusting the position of the already constructed map elements is achieved, and the difficulty of manual input and map maintenance is reduced.

Description

Technical Field

[0001] This invention relates to the technical field of high-precision map construction, and more specifically, to a method and system for eliminating deviations in voxel-based point cloud high-precision map fusion. Background Technology

[0002] High-precision maps, as an essential auxiliary tool for autonomous driving, rely heavily on high mapping accuracy. However, large-scale mapping often requires fusing multiple data segments. These segments contain overlapping mapping routes. Generally, due to sensor errors and environmental factors, the coordinates of points at the same location will deviate to some extent after multiple data collections, affecting mapping accuracy. Data fusion eliminates these inconsistencies, but the poses of the data before and after fusion change. Mappers then need to modify the map based on these pose changes, increasing both cost and time.

[0003] A patent document with publication number CN115526914A discloses a multi-sensor-based method for real-time robot localization and color map fusion mapping. This method rapidly collects information about the surrounding environment and constructs a global color map in real time. Vision, laser, and inertial sensor acquisition modules read sensor information, accumulate laser point cloud data to form point cloud scan frames, and then perform preprocessing to obtain extracted feature point clouds. Simultaneously, the data collected by different sensors are synchronized in time. The acquired laser and inertial data are used to perform real-time self-localization of the robot, while simultaneously constructing a single-frame point cloud map. RGB information collected by the vision sensor is used to render the texture and color of the constructed global point cloud map, creating a single-frame color map.

[0004] Regarding the aforementioned technologies, the inventors believe that in existing technologies, mapmakers must modify maps created from the original mapping data based on pose changes after data fusion, increasing mapping costs and time. Therefore, a new technical solution is needed to improve these technical problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for eliminating deviations in voxel-based point cloud high-precision map fusion.

[0006] In a first aspect, the present invention provides a method for eliminating deviations in the fusion of high-precision point cloud maps based on voxels, comprising the following steps:

[0007] Step S1: Collect point cloud data of the map A to be constructed, and obtain point cloud dataset A;

[0008] Step S2: Generate map A by creating a map from the point cloud dataset A based on the poses before fusion;

[0009] Step S3: Collect point cloud data of the map B to be constructed to obtain point cloud dataset B; and since there is a line overlap between map A and map B, merge point cloud dataset A and point cloud dataset B to generate the merged pose.

[0010] Step S4: Generate map B by using the point cloud dataset B based on the fused pose;

[0011] Step S5: Calculate the pose change for point cloud dataset A and the poses before and after fusion, and generate a voxel description file based on the voxel coordinates;

[0012] Step S6: The point cloud dataset A is adjusted by querying the voxel description file. Based on the adjusted point cloud dataset A and the fused pose, the map A is adjusted to obtain map A'.

[0013] Step S7: Overlay and merge map A' and map B to obtain map C.

[0014] Preferably, in step S1, point cloud data of map A is collected using LiDAR, and the pose before fusion is obtained through SLAM mapping optimization.

[0015] Preferably, in step S3, after point cloud dataset A and point cloud dataset B are fused, SLAM mapping is used to stitch and optimize map A and map B to obtain the fused pose.

[0016] Preferably, the point cloud dataset B in step S4 includes a portion that is fused with the point cloud dataset A and a portion that is not fused with the point cloud dataset A.

[0017] Preferably, the point cloud dataset A in step S5 includes a portion fused with point cloud dataset B and a portion not fused with point cloud dataset B.

[0018] Preferably, step S5 includes:

[0019] Step S5.1: For each frame of point cloud in point cloud dataset A, project the point cloud according to the pose before fusion and the pose after fusion respectively;

[0020] Step S5.2: Calculate the projection coordinate offset based on the two projection results;

[0021] Step S5.3: Calculate the voxel coordinates based on the UTM coordinates of each frame of point cloud and store the offset in the voxel;

[0022] Step S5.4: Traverse all frame point clouds, count the offset in each voxel, and calculate the average offset.

[0023] Step S5.5: Expand the scope of the point cloud computing, calculate the offset of voxels that are not covered by the point cloud, and finally save all voxels and their corresponding offsets into a voxel description file.

[0024] Preferably, in step S5.5, the voxel description file undergoes corresponding version iterations based on the map version iterations. The point cloud dataset A queries the version of the voxel description file for adjustment. The version relationship is used to determine whether the version of the map before adjustment corresponds to the version of the voxel description file. If it matches, the voxel description file is used for map adjustment; otherwise, the voxel description file is not used for map adjustment.

[0025] Secondly, the present invention also provides a voxel-based point cloud high-precision map fusion deviation elimination system, comprising the following modules:

[0026] Module M1: Collects point cloud data of the map A to be built, and obtains point cloud dataset A;

[0027] Module M2: Generates map A by creating a map from point cloud dataset A based on the poses before fusion;

[0028] Module M3: Collects point cloud data of the map B to be constructed, and obtains point cloud dataset B; since there is a line overlap between map A and map B, the point cloud dataset A and point cloud dataset B are fused to generate the fused pose;

[0029] Module M4: Generates map B by using the fused point cloud dataset B to create a map.

[0030] Module M5: Calculates pose changes for point cloud dataset A and the poses before and after fusion, and generates voxel description files based on voxel coordinates;

[0031] Module M6: Point cloud dataset A is adjusted by querying the voxel description file. Based on the adjusted point cloud dataset A and the fused pose, map A is adjusted to obtain map A'.

[0032] Module M7: Overlays and merges map A' and map B to obtain map C.

[0033] Preferably, the module M1 uses a lidar to collect point cloud data of map A and obtains the pose before fusion through SLAM mapping optimization.

[0034] Preferably, in module M3, after point cloud dataset A and point cloud dataset B are fused, SLAM mapping is used to stitch and optimize map A and map B to obtain the fused pose.

[0035] Preferably, the point cloud dataset B in module M4 includes a portion that is fused with point cloud dataset A and a portion that is not fused with point cloud dataset A.

[0036] Preferably, the point cloud dataset A in module M5 includes a portion fused with point cloud dataset B and a portion not fused with point cloud dataset B.

[0037] Preferably, the module M5 includes:

[0038] Module M5.1: For each frame of point cloud in point cloud dataset A, project the point cloud according to the pose before fusion and the pose after fusion respectively;

[0039] Module M5.2: Calculate the projection coordinate offset based on the two projection results;

[0040] Module M5.3: Calculates voxel coordinates based on the UTM coordinates of each frame of point cloud and stores the offset in the voxel;

[0041] Module M5.4: Traverse all frame point clouds, count the offset in each voxel, and calculate the average offset.

[0042] Module M5.5: Extends the scope of point cloud computing, calculates the offset of voxels that are not covered by the point cloud, and finally saves all voxels and their corresponding offsets into a voxel description file.

[0043] Preferably, in module M5.5, the voxel description file undergoes corresponding version iterations based on the map version iterations. The point cloud dataset A queries the version of the voxel description file for adjustment. The version relationship is used to determine whether the version of the map before adjustment corresponds to the version of the voxel description file. If it matches, the voxel description file is used for map adjustment; otherwise, the voxel description file is not used for map adjustment.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. Based on the pose changes before and after point cloud fusion, this invention describes the changes in point cloud through voxels. The points in the high-precision map can be automatically adjusted by querying the offset in the voxels. In this way, it is not necessary to repeatedly modify the already built high-precision map, which can greatly shorten the mapping cost and cycle of high-precision map.

[0046] 2. This invention generates a high-precision map file that can be adjusted based on the poses of the map data before and after fusion. By using this file, the positions of the already built map elements can be automatically adjusted, reducing manual input and the difficulty of map maintenance.

[0047] 3. The input of this invention is the point cloud dataset obtained by the lidar sensor and the poses of the point cloud before and after fusion. The high-precision map is automatically adjusted by querying the voxel description file. Attached Figure Description

[0048] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0049] Figure 1 This is a flowchart of the process of the present invention;

[0050] Figure 2 This is a schematic diagram of the workflow of one embodiment of the present invention;

[0051] Figure 3 This is a flowchart of the voxel description file generation process of the present invention;

[0052] Figure 4 This invention provides a point cloud map that integrates the previous and subsequent poses;

[0053] Figure 5 This is an overlay display of the point cloud map after fusion and the point cloud map after offsetting the voxel description file read from the point cloud before fusion, according to the present invention. Detailed Implementation

[0054] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0055] According to the present invention, a voxel-based point cloud high-precision map fusion deviation elimination method is provided, such as... Figure 1 and Figure 2 The process includes the following steps:

[0056] Step S1: Collect point cloud data of the map A to be constructed to obtain point cloud dataset A; use LiDAR to collect point cloud data of map A, and use SLAM mapping optimization to obtain the pose before fusion.

[0057] Step S2: Generate a map A from the point cloud dataset A based on the poses before fusion.

[0058] Step S3: Collect point cloud data of the map B to be constructed to obtain point cloud dataset B; since there is a line overlap between map A and map B, merge point cloud dataset A and point cloud dataset B to generate the merged pose; after merging point cloud dataset A and point cloud dataset B, use SLAM mapping to stitch and optimize map A and map B to obtain the merged pose.

[0059] Step S4: Create a map from point cloud dataset B based on the fused pose; point cloud dataset B includes the part fused with point cloud dataset A and the part not fused with point cloud dataset A.

[0060] Step S5: Calculate the pose change for point cloud dataset A and the poses before and after fusion, and generate a voxel description file based on the voxel coordinates; point cloud dataset A includes the part fused with point cloud dataset B and the part not fused with point cloud dataset B.

[0061] In one embodiment, step S5 includes:

[0062] Step S5.1: For each frame of point cloud in point cloud dataset A, project the point cloud according to the pose before fusion and the pose after fusion respectively;

[0063] Step S5.2: Calculate the projection coordinate offset based on the two projection results;

[0064] Step S5.3: Calculate the voxel coordinates based on the UTM coordinates of each frame of point cloud and store the offset in the voxel;

[0065] Step S5.4: Traverse all frame point clouds, count the offset in each voxel, and calculate the average offset.

[0066] Step S5.5: Expand the scope of the point cloud computing, calculate the offset of voxels that are not covered by the point cloud, and finally save all voxels and their corresponding offsets into a voxel description file.

[0067] Furthermore, in step S5.5, the voxel description file undergoes corresponding version iterations based on the map version iterations. The point cloud dataset A queries the version of the voxel description file for adjustment. The version relationship is used to determine whether the version of the map before adjustment corresponds to the version of the voxel description file. If it matches, the voxel description file is used for map adjustment; otherwise, the voxel description file is not used for map adjustment.

[0068] Step S6: The point cloud dataset A is adjusted by querying the voxel description file. Based on the adjusted point cloud dataset A and the fused pose, the map A is adjusted to obtain map A'.

[0069] Step S7: Overlay and merge map A' and map B to obtain map C.

[0070] The present invention also provides a voxel-based point cloud high-precision map fusion deviation elimination system. The voxel-based point cloud high-precision map fusion deviation elimination system can be implemented by executing the process steps of the voxel-based point cloud high-precision map fusion deviation elimination method. That is, those skilled in the art can understand the voxel-based point cloud high-precision map fusion deviation elimination method as a preferred embodiment of the voxel-based point cloud high-precision map fusion deviation elimination system.

[0071] This invention also provides a voxel-based point cloud high-precision map fusion deviation elimination system, comprising the following modules:

[0072] Module M1: Collects point cloud data of the map A to be built, and obtains point cloud dataset A; uses LiDAR to collect point cloud data of map A, and obtains the pose before fusion through SLAM mapping optimization.

[0073] Module M2: Generates map A by taking the point cloud dataset A based on the pose before fusion.

[0074] Module M3: Collects point cloud data of the map B to be constructed, and obtains point cloud dataset B; since there is a line overlap between map A and map B, point cloud dataset A and point cloud dataset B are merged to generate the merged pose; after the point cloud dataset A and point cloud dataset B are merged, SLAM mapping is used to stitch and optimize map A and map B to obtain the merged pose.

[0075] Module M4: Creates a map from point cloud dataset B based on the fused pose; point cloud dataset B includes the part fused with point cloud dataset A and the part not fused with point cloud dataset A.

[0076] Module M5: Calculates pose changes for point cloud dataset A and the poses before and after fusion, and generates voxel description files based on voxel coordinates; point cloud dataset A includes the part fused with point cloud dataset B and the part not fused with point cloud dataset B.

[0077] In one embodiment, module M5 includes:

[0078] Module M5.1: For each frame of point cloud in point cloud dataset A, project the point cloud according to the pose before fusion and the pose after fusion respectively;

[0079] Module M5.2: Calculate the projection coordinate offset based on the two projection results;

[0080] Module M5.3: Calculates voxel coordinates based on the UTM coordinates of each frame of point cloud and stores the offset in the voxel;

[0081] Module M5.4: Traverse all frame point clouds, count the offset in each voxel, and calculate the average offset.

[0082] Module M5.5: Extends the scope of point cloud computing, calculates the offset of voxels that are not covered by the point cloud, and finally saves all voxels and their corresponding offsets into a voxel description file.

[0083] Furthermore, in module M5.5, the voxel description file undergoes corresponding version iterations based on the map version iterations. The point cloud dataset A queries the version of the voxel description file for adjustment. The version relationship is used to determine whether the version of the map before adjustment corresponds to the version of the voxel description file. If it matches, the voxel description file is used for map adjustment; otherwise, the voxel description file is not used for map adjustment.

[0084] Module M6: The point cloud dataset A is adjusted by querying the voxel description file. Based on the adjusted point cloud dataset A and the fused pose, the map A is adjusted to obtain map A'.

[0085] Module M7: Overlays and merges map A' and map B to obtain map C.

[0086] In one embodiment, the core of the present invention is to use voxels to describe the pose changes before and after point cloud fusion, and to automatically adjust the high-precision map points that need to change positions by querying, thereby ensuring the accuracy and uniqueness of the high-precision map.

[0087] The core of this invention lies in defining high-precision map version iterations to ensure that one map corresponds to one voxel file reflecting the pose changes of point cloud fusion.

[0088] The design process of this invention is as follows: First, the point cloud dataset A acquired by the LiDAR for the first time is used to create a map based on the pose before fusion, generating map A. If a new point cloud dataset B is acquired at this time, and there is a line overlap between point cloud datasets A and B, the mapping will re-fuse point cloud datasets A and B to generate a fused pose. The point cloud dataset B is then used to create a map based on the fused pose, generating map B. At this time, point cloud dataset B includes fused and non-fused parts. Since the pose of point cloud dataset A has changed due to point cloud fusion, the pose of the non-fused part of point cloud dataset A will also change. Map a adjusts the entire map, so map A created based on point cloud dataset A also needs to be adjusted. The point cloud dataset A and the poses before and after fusion are input into this invention, and a voxel description file is output. The points of map A are translated by querying the voxels corresponding to the file, thereby completing the automatic adjustment of map A' based on point cloud dataset A and the corresponding fused pose. Then, map A' and map B are simply superimposed to obtain the fused map C.

[0089] The input to this invention is a point cloud dataset obtained from a LiDAR sensor and the corresponding poses before and after point cloud fusion. Due to sensor errors, each time new data is added to generate a map, it needs to be re-stitched and optimized. The pose before optimization is the pose generated by SLAM mapping and stitching optimization without adding new data. The pose after optimization is the pose obtained by re-stitching and optimizing the map after adding new data. First, each frame of the point cloud dataset needs to be projected according to the poses before and after fusion. Then, the coordinate offset is calculated based on the results of the two projections, and the voxel coordinates are calculated based on the UTM coordinates of the points before fusion. The calculated offset is stored in the voxel. After traversing all the point clouds, the offsets in each voxel are statistically analyzed, and the average offset is calculated. Finally, the calculation range of the point cloud is expanded, and the offset values ​​of the voxels are calculated to include the offsets of voxels not covered by the point cloud. All voxels and their corresponding offsets are saved as a voxel description file for use in high-precision map adjustments. Each part will be described in detail below.

[0090] This invention uses the timestamp of a lidar point cloud to determine the location of point P in the vehicle coordinate system within the point cloud. b Calculate the UTM coordinates of the points before and after fusion. The coordinate offset is calculated for points in the point cloud, and then the coordinate offset value is calculated.

[0091]

[0092]

[0093] P Δ =P u -Pm

[0094] in, Let be the homogeneous coordinates of a point in the vehicle coordinate system. Calculate the coordinate offset for a point in the point cloud. T is the transformation matrix from the vehicle coordinate system to the UTM coordinate system before fusion. T represents the homogeneous coordinates in the UTM coordinate system before fusion. m This is the transformation matrix from the merged vehicle coordinate system to the UTM coordinate system. P represents the homogeneous coordinates in the merged UTM coordinate system. Δ This represents the UTM coordinate offset value before and after the point is merged.

[0095] This invention, after obtaining the coordinate offset values ​​of a point in a point cloud frame, calculates the voxel coordinates corresponding to the point's coordinates in the UTM coordinate system before fusion. Calculating the voxel first requires determining the voxel size. Generally, the smaller the voxel, the closer the adjusted high-precision map is to the high-precision map created based on the fused pose. However, excessively small voxels can lead to a cubic increase in computational load. A suitable voxel size can meet the accuracy requirements of the high-precision map while keeping the computational load within a reasonable range. The voxel size is defined arbitrarily, and experiments have shown that a relatively optimal value is within 0.1 to 1. After determining the voxel size, the voxel coordinates corresponding to the point can be obtained by dividing the point's UTM coordinates before fusion by the voxel size and then rounding.

[0096] v x =[u x / r x ]

[0097] v y =[u y / r y ]

[0098] v z =[u z / r z ]

[0099] Among them, v x v y and v z For voxel coordinates, u x u y and u z For P u The coordinate value in r x r y and r z is a defined resolution constant, representing the size of the voxel, and [] is the rounding operation.

[0100] After calculating the coordinate offset values ​​of voxels within a frame of point cloud, the voxel coverage of this frame needs to be expanded. During the high-precision map adjustment phase, considering special cases, such as a wide road in the high-precision map where the mapping point cloud only covers a portion of the map, the pose adjustments before and after fusion will cause offsets in the covered areas of the point cloud. Roads in adjacent mapping point clouds that are not covered will not be adjusted, leading to inconsistencies in local areas of the high-precision map. Therefore, it is necessary to expand the voxel calculation range, calculating offset values ​​for voxels outside the mapping point cloud's coverage area as well.

[0101] Since there are no physical points in the extended region, it is first necessary to limit the extension range of the point cloud in each frame, and then to virtualize a point within the voxel to calculate the coordinate offset of the virtual point.

[0102]

[0103]

[0104]

[0105] Where, u′ x ,u′ y and u′ z The UTM coordinates of the center point of the extended voxel V′.

[0106] After obtaining the UTM coordinates of the extended voxel center point, the corresponding fused pose information is found, and the coordinate offset of the virtual point is calculated.

[0107]

[0108] P′ Δ =P′ u -P′ m

[0109] in, The homogeneous coordinates of the virtual center point within the voxel in the UTM coordinate system before fusion. P′ represents the homogeneous coordinates of the virtual center point within the voxel in the UTM coordinate system after fusion. Δ The UTM coordinate offset values ​​of virtual points within the voxel before and after fusion.

[0110] Since a voxel may contain multiple points in the same frame point cloud and multiple points in different frames, or may be covered multiple times during the expansion of the voxel range in different frame point clouds, and the phenomenon of point cloud displacement caused by pose changes before and after fusion is consistent within a certain range in the same point cloud dataset, the average offset value of all points within the voxel range can be obtained by averaging the offset values ​​of the points before and after fusion within the voxel range. The calculated average voxel offset refers to the offset of a point within a single voxel.

[0111]

[0112] Where G is the mean coordinate offset of points within the voxel range, n is the number of points within the voxel range, and P Δi Let be the pose coordinate offset of the i-th point before and after fusion, i∈[0,n-1].

[0113] Finally, it is necessary to confirm the version relationship. The version relationship requires describing the map versions before and after fusion. One voxel description file corresponds to one version map. The version relationship is used to determine whether the version of the high-precision map before adjustment corresponds to the version of the voxel description file. If they do not meet the requirements, the voxel description file should not be used to adjust the high-precision map, otherwise an error will occur.

[0114] The input to this invention is a point cloud dataset obtained from a lidar sensor and the corresponding poses before and after point cloud fusion. Figure 4 As shown, the point cloud maps with poses superimposed before and after fusion are shown. The dense point cloud is the point cloud map with poses superimposed before fusion, and the sparse point cloud is the point cloud map with poses superimposed after fusion. It can be seen that the pose shift of the point cloud before and after fusion is large, and the ghosting phenomenon is obvious. Figure 5 The image shows a superimposed view of the point cloud map after fusion and the point cloud map after offsetting the voxel description file before fusion. It can be seen that the two point clouds are almost completely overlapping, which shows that the high-precision map can be automatically adjusted by querying the voxel description file.

[0115] Those skilled in the art can understand this embodiment as a more specific description of the embodiments.

[0116] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0117] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for eliminating bias in voxel-based point cloud high-precision map fusion, characterized in that, Includes the following steps: Step S1: Collect point cloud data of the map A to be constructed, and obtain point cloud dataset A; Step S2: Generate map A by creating a map from the point cloud dataset A based on the poses before fusion; Step S3: Collect point cloud data of the map B to be constructed to obtain point cloud dataset B; and since there is a line overlap between map A and map B, merge point cloud dataset A and point cloud dataset B to generate the merged pose. Step S4: Generate map B by using the point cloud dataset B based on the fused pose; Step S5: Calculate the pose change for point cloud dataset A and the poses before and after fusion, and generate a voxel description file based on the voxel coordinates; Step S6: The point cloud dataset A is adjusted by querying the voxel description file. Based on the adjusted point cloud dataset A and the fused pose, the map A is adjusted to obtain map A'. Step S7: Overlay and merge map A' and map B to obtain map C; Step S5 includes: Step S5.1: For each frame of point cloud in point cloud dataset A, project the point cloud according to the pose before fusion and the pose after fusion respectively; Step S5.2: Calculate the projection coordinate offset based on the two projection results; Step S5.3: Calculate the voxel coordinates based on the UTM coordinates of each frame of point cloud and store the offset in the voxel; Step S5.4: Traverse all frame point clouds, count the offset in each voxel, and calculate the average offset. Step S5.5: Expand the scope of the point cloud computing, calculate the offset of voxels that are not covered by the point cloud, and finally save all voxels and their corresponding offsets into a voxel description file.

2. The method for eliminating deviations in voxel-based point cloud high-precision map fusion according to claim 1, characterized in that, In step S1, point cloud data of map A is collected using LiDAR, and the pose before fusion is obtained through SLAM mapping optimization.

3. The method for eliminating deviations in voxel-based point cloud high-precision map fusion according to claim 1, characterized in that, In step S3, after point cloud dataset A and point cloud dataset B are fused, SLAM mapping is used to stitch and optimize map A and map B to obtain the fused pose.

4. The method for eliminating deviations in voxel-based point cloud high-precision map fusion according to claim 1, characterized in that, The point cloud dataset B in step S4 includes the portion that is fused with point cloud dataset A and the portion that is not fused with point cloud dataset A.

5. The method for eliminating deviations in voxel-based point cloud high-precision map fusion according to claim 1, characterized in that, The point cloud dataset A in step S5 includes a portion that is fused with point cloud dataset B and a portion that is not fused with point cloud dataset B.

6. The method for eliminating deviations in voxel-based point cloud high-precision map fusion according to claim 1, characterized in that, In step S5.5, the voxel description file undergoes corresponding version iterations based on the map version iterations. The point cloud dataset A queries the version of the voxel description file for adjustment. The version relationship is used to determine whether the version of the map before adjustment corresponds to the version of the voxel description file. If it matches, the voxel description file is used for map adjustment; otherwise, the voxel description file is not used for map adjustment.

7. A voxel-based point cloud high-precision map fusion deviation elimination system, characterized in that, Includes the following modules: Module M1: Collects point cloud data of the map A to be built, and obtains point cloud dataset A; Module M2: Generates map A by creating a map from point cloud dataset A based on the poses before fusion; Module M3: Collects point cloud data of the map B to be constructed, and obtains point cloud dataset B; since there is a line overlap between map A and map B, the point cloud dataset A and point cloud dataset B are fused to generate the fused pose; Module M4: Generates map B by using the fused point cloud dataset B to create a map. Module M5: Calculates pose changes for point cloud dataset A and the poses before and after fusion, and generates voxel description files based on voxel coordinates; Module M6: Point cloud dataset A is adjusted by querying the voxel description file. Based on the adjusted point cloud dataset A and the fused pose, map A is adjusted to obtain map A'. Module M7: Overlays and merges map A' and map B to obtain map C; The module M5 includes: Module M5.1: For each frame of point cloud in point cloud dataset A, project the point cloud according to the pose before fusion and the pose after fusion respectively; Module M5.2: Calculate the projection coordinate offset based on the two projection results; Module M5.3: Calculates voxel coordinates based on the UTM coordinates of each frame of point cloud and stores the offset in the voxel; Module M5.4: Traverse all frame point clouds, count the offset in each voxel, and calculate the average offset. Module M5.5: Extends the scope of point cloud computing, calculates the offset of voxels that are not covered by the point cloud, and finally saves all voxels and their corresponding offsets into a voxel description file.

8. The voxel-based point cloud high-precision map fusion deviation elimination system according to claim 7, characterized in that, The module M1 uses a lidar to collect point cloud data from map A and then uses SLAM mapping optimization to obtain the pose before fusion.

9. The voxel-based point cloud high-precision map fusion deviation elimination system according to claim 8, characterized in that, In module M3, point cloud datasets A and B are fused, and then SLAM mapping is used to stitch and optimize maps A and B to obtain the fused pose.

10. The voxel-based point cloud high-precision map fusion deviation elimination system according to claim 9, characterized in that, The point cloud dataset B in module M4 includes a portion that is fused with point cloud dataset A and a portion that is not fused with point cloud dataset A.

11. The voxel-based point cloud high-precision map fusion deviation elimination system according to claim 10, characterized in that, The point cloud dataset A in module M5 includes a portion that is fused with point cloud dataset B and a portion that is not fused with point cloud dataset B.

12. The voxel-based point cloud high-precision map fusion deviation elimination system according to claim 7, characterized in that, In module M5.5, the voxel description file undergoes corresponding version iterations based on the map version. The point cloud dataset A queries the version of the voxel description file for adjustment. The version relationship is used to determine whether the version of the map before adjustment corresponds to the version of the voxel description file. If it matches, the voxel description file is used for map adjustment; otherwise, the voxel description file is not used for map adjustment.

Images