High-precision map construction method and device, storage medium and equipment
By dividing the LiDAR data point cloud set into ground points, corner points, and surface points, calculating the distances of various point clouds, and introducing ground points into loop closure matching, the problem of excessively rapid z-axis divergence in high-precision map construction is solved, thus improving matching accuracy and efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIUZHI (SUZHOU) INTELLIGENT TECH CO LTD
- Filing Date
- 2023-02-28
- Publication Date
- 2026-05-19
AI Technical Summary
In the process of building high-precision maps, the problem of excessively rapid divergence of the z-axis, especially in traditional surveying methods and existing SLAM technology, is that insufficient constraints on the z-axis lead to large cumulative errors, affecting the accuracy of the map and computational efficiency.
The lidar data point cloud set is divided into ground points, corner points, and surface points. The distance between each type of point cloud is calculated. The z-axis constraint is added by processing the ground points separately, and the ground points are introduced in the loop closure matching to decouple the 6-DOF pose estimation. The pose is optimized using ICP and CSM algorithms.
It improves the matching success rate and accuracy under large cumulative errors, shortens the calculation time, alleviates the problem of excessively rapid divergence of the z-axis, and enhances the efficiency of high-precision map construction.
Smart Images
Figure CN116299542B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of image processing technology, and in particular to a method, apparatus, storage medium and device for constructing high-precision maps. Background Technology
[0002] In autonomous driving solutions that rely on high-precision maps, high-precision maps provide accurate prior environmental information for localization, perception, and PnC (Planning and Control). They have advantages such as high precision, multiple elements, and high "freshness," and are an important prerequisite for vehicles to make safe decisions and judgments. They are also an important guarantee for autonomous vehicles to complete various tasks safely and efficiently. Therefore, building high-precision maps is an important foundation and key link in autonomous driving solutions.
[0003] There are generally two methods for constructing high-precision maps: one is to create high-precision maps using traditional surveying methods, specifically by using high-precision inertial navigation equipment to collect the vehicle's pose and projecting the onboard LiDAR data to generate a high-precision map. The other is to create high-precision maps using SLAM (Simultaneous Localization and Mapping) technology, specifically by using registration methods to estimate the pose corresponding to each frame of onboard LiDAR data, eliminating errors through loop closure detection and global optimization, and finally generating a high-precision map.
[0004] Traditional mapping methods, while capable of quickly measuring vehicle pose, are susceptible to environmental interference, particularly between buildings and under dense trees, often resulting in pose errors that fail to meet the demands of high-precision maps. In situations where high-precision inertial navigation equipment cannot acquire pose data, SLAM (Simultaneous Localization and Mapping) technology can be used for accurate pose estimation. SLAM technology includes a front-end odometry system. Front-end odometry can employ methods such as ICP (Iterative Closest Point), GICP (Generalized Iterative Closest Point), and NDT (Normal Distribution Transform), but these methods are computationally intensive, and their efficiency decreases significantly as the map area increases. Therefore, feature-based methods have gradually become the mainstream approach for front-end odometry, such as LOAM (Lidar Odometry and Mapping). However, LOAM only extracts corner points and polygon points in the environment as features for pose estimation, resulting in fewer constraints on the z-axis. The cumulative error increases with the distance traveled, causing rapid divergence along the z-axis. Summary of the Invention
[0005] This application provides a method, apparatus, storage medium, and device for constructing high-precision maps, to solve the problem of excessively rapid z-axis divergence during high-precision mapping. The technical solution is as follows:
[0006] On the one hand, a method for constructing a high-precision map is provided, the method comprising:
[0007] Obtain the point cloud set of the current frame and the previous frame of LiDAR data, wherein the point cloud set includes ground points, corner points and surface points;
[0008] Obtain a point cloud distance model, and calculate the distance between the current frame and the same type of point cloud in the previous frame based on the point cloud distance model;
[0009] An error function is created based on the distance, and the pose of the current frame relative to the previous frame is calculated based on the error function.
[0010] Create a high-precision map based on the pose.
[0011] In one possible implementation, obtaining the point cloud distance model and calculating the distance between similar point clouds in the current frame and the previous frame based on the point cloud distance model includes:
[0012] When the point cloud distance model is a ground point distance model, the planar distance from each ground point in the point cloud set of the current frame to the ground in the previous frame is calculated, and the ground is composed of ground points in the point cloud set of the previous frame.
[0013] In one possible implementation, obtaining the point cloud distance model and calculating the distance between similar point clouds in the current frame and the previous frame based on the point cloud distance model includes:
[0014] When the point cloud distance model is a corner distance model, the corner distance from each corner point in the point cloud set of the current frame to the corner point of the previous frame is calculated.
[0015] In one possible implementation, obtaining the point cloud distance model and calculating the distance between similar point clouds in the current frame and the previous frame based on the point cloud distance model includes:
[0016] When the point cloud distance model is a face distance model, the planar distance from each face in the point cloud set of the current frame to the plane in the previous frame is calculated, and the plane is composed of face points in the point cloud set of the previous frame.
[0017] In one possible implementation, the method further includes:
[0018] Perform loop closure detection on the point cloud set of each frame of LiDAR data;
[0019] If a loop is detected between the m-th point cloud set and the n-th point cloud set, then the ground points in the m-th point cloud set and the n-th point cloud set are matched to obtain the transformation values of the z-axis, roll angle and pitch angle between the lidar data of the m-th frame and the n-th frame.
[0020] Using the transformed values of the z-axis, roll angle, and pitch angle as initial values, calculate the transformed values of the x-axis, y-axis, and yaw angle between the LiDAR data of the m-th frame and the n-th frame;
[0021] Using the transformed values of the x-axis, y-axis, z-axis, yaw angle, roll angle, and pitch angle as initial values, the ICP algorithm is used to calculate the final transformed value between the LiDAR data of the m-th frame and the n-th frame.
[0022] In one possible implementation, the step of using the transformed values of the z-axis, roll angle, and pitch angle as initial values to calculate the transformed values of the x-axis, y-axis, and yaw angle between the lidar data of the m-th frame and the n-th frame includes:
[0023] Using the transformed values of the z-axis, roll angle, and pitch angle as initial values, the corner points and face points in the m-th point cloud set and the n-th point cloud set are compressed into a two-dimensional map;
[0024] The CSM scanning matching algorithm is used to match the two-dimensional map to obtain the transformation values of the x-axis, y-axis and yaw angle between the lidar data of the m-th frame and the n-th frame.
[0025] In one possible implementation, creating a high-precision map based on the pose includes:
[0026] The high-precision map is created based on the pose and the final transformation value.
[0027] On the one hand, a high-precision map construction apparatus is provided, the apparatus comprising:
[0028] The acquisition module is used to acquire the point cloud set of the current frame and the previous frame of LiDAR data, wherein the point cloud set includes ground points, corner points and surface points;
[0029] The acquisition module is further configured to acquire a point cloud distance model and calculate the distance between the current frame and the same type of point cloud in the previous frame based on the point cloud distance model.
[0030] The calculation module is used to create an error function based on the distance and calculate the pose of the current frame relative to the previous frame based on the error function.
[0031] A module is created to generate a high-precision map based on the given pose.
[0032] On the one hand, a computer-readable storage medium is provided, wherein at least one instruction is stored therein, the at least one instruction being loaded and executed by a processor to implement the method for constructing a high-precision map as described above.
[0033] On one hand, a computer device is provided, the computer device including a processor and a memory, the memory storing at least one instruction, the instruction being loaded and executed by the processor to implement the high-precision map construction method as described above.
[0034] The beneficial effects of the technical solution provided in this application include at least the following:
[0035] By dividing the point cloud into three categories—ground points, corner points, and polygon points—and then creating a high-precision map, we can increase the diversity of features and also increase the constraint on the z-axis by processing the ground points separately, thus alleviating the problem of excessively rapid z-axis divergence.
[0036] By incorporating ground points into the loop matching process, the estimation of the 6-DOF pose can be decoupled, achieving a higher matching success rate and accuracy even with larger cumulative errors, while also shortening the computation time. Attached Figure Description
[0037] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0038] Figure 1 This is a flowchart of a method for constructing a high-precision map according to an embodiment of this application;
[0039] Figure 2 This is a flowchart of a method for constructing a high-precision map according to an embodiment of this application;
[0040] Figure 3 This is a structural block diagram of a high-precision map construction apparatus provided in another embodiment of this application;
[0041] Figure 4 This is a structural block diagram of a high-precision map construction apparatus provided in another embodiment of this application. Detailed Implementation
[0042] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.
[0043] Please refer to Figure 1 This document illustrates a flowchart of a method for constructing a high-precision map according to an embodiment of this application. This high-precision map construction method can be applied to a computer device. The high-precision map construction method may include:
[0044] Step 101: Obtain the point cloud set of the current frame and the previous frame of LiDAR data. The point cloud set includes ground points, corner points and surface points.
[0045] Autonomous vehicles are equipped with LiDAR, which can periodically collect data from the surrounding environment and generate a point cloud set based on each frame of LiDAR data.
[0046] After acquiring a point cloud set, the computer device preprocesses the point cloud, dividing it into ground points and non-ground points. There are many methods for extracting ground points from the point cloud, and this embodiment does not limit the methods. For non-ground points, the computer device can divide them into corner points and face points based on curvature. Therefore, a point cloud set includes ground points, corner points, and face points.
[0047] Step 102: Obtain the point cloud distance model and calculate the distance between the same type of point cloud in the current frame and the previous frame based on the point cloud distance model.
[0048] Computer devices can model point clouds based on the characteristics of each category to obtain point cloud distance models, and use these models to calculate the distance between point clouds of the same type in the current frame and the previous frame. In this embodiment, the point cloud distance model created for ground points is called the ground point distance model, the point cloud distance model created for corner points is called the corner distance model, and the point cloud distance model created for area points is called the area distance model. The calculation methods for the three point cloud distance models are explained below.
[0049] (1) The point cloud distance model is the ground point distance model;
[0050] Specifically, obtaining a point cloud distance model and calculating the distance between similar point clouds in the current frame and the previous frame based on the point cloud distance model can include: when the point cloud distance model is a ground point distance model, calculating the planar distance from each ground point in the point cloud set of the current frame to the ground in the previous frame, where the ground is composed of ground points in the point cloud set of the previous frame. The ground point distance model is used to estimate the planar distance from the current ground point to the ground points in the target point cloud set.
[0051] Assume the current frame is the (k+1)th frame, and its corresponding point cloud set is X. k+1 The previous frame is the k-th frame, and its corresponding point cloud set is X. kFor a ground point i, where i∈X k+1 Find X by searching for its nearest neighbor. k Given three nearby ground points j, l, and p, and these three points j, l, and p can form a ground, the planar distance from the current ground point i to this ground is...
[0052] (2) The point cloud distance model is a corner distance model;
[0053] Specifically, obtaining the point cloud distance model and calculating the distance between similar point clouds in the current frame and the previous frame based on the point cloud distance model can include: when the point cloud distance model is a corner distance model, calculating the angular distance from each corner point in the point cloud set of the current frame to a corner point in the previous frame. The corner distance model is used to estimate the angular distance from the current corner point to the target point cloud.
[0054] Assume the current frame is the (k+1)th frame, and its corresponding point cloud set is X. k+1 The previous frame is the k-th frame, and its corresponding point cloud set is X. k For a corner point i, where i∈X k+1 Find X by searching for nearest neighbors k Given two adjacent corner points j and l, and these two corner points j and l can form a straight line, the distance from the current corner point i to the angle of the connecting line is...
[0055] (3) The point cloud distance model is a surface-to-point distance model;
[0056] Specifically, obtaining a point cloud distance model and calculating the distance between similar point clouds in the current frame and the previous frame based on the point cloud distance model can include: when the point cloud distance model is a face-to-point distance model, calculating the planar distance from each face point in the current frame's point cloud set to a plane in the previous frame, where the plane is composed of face points in the previous frame's point cloud set. The face-to-point distance model is used to estimate the planar distance from the current face point to the face points in the target point cloud set.
[0057] Assume the current frame is the (k+1)th frame, and its corresponding point cloud set is X. k+1 The previous frame is the k-th frame, and its corresponding point cloud set is X. k For a point i, where i∈X k+1 Find X by searching for nearest neighbors k Given three points j, l, and p that are close to each other, and these three points j, l, and p can form a plane, the planar distance from the current point i to this plane is...
[0058] Step 103: Create an error function based on the distance, and calculate the pose of the current frame relative to the previous frame based on the error function.
[0059] Calculating the distance between similar point clouds in the current frame and the previous frame can also be called motion estimation, the purpose of which is to calculate the transformed pose.
[0060] The computer device constructs a transformation relationship f(*) for point clouds with different features, such that:
[0061] f c (X (k+1，i) T k+1 )=d c , i∈ε k+1
[0062]
[0063] f g (X (k+1，i) T k+1 )=d g , i∈δ k+1
[0064] Where, ε k+1 This represents the set of corner points in the point cloud of the current frame. δ represents the set of facets in the point cloud set of the current frame. k+1 This represents the set of ground points in the point cloud set of the current frame.
[0065] Computer devices can construct the error function f(T) k+1 The optimal pose transformation is obtained by minimizing d to 0 using the Levenberg-Marquardt (LM) method.
[0066] Step 104: Create a high-precision map based on the pose.
[0067] After obtaining the pose, the computer equipment can eliminate errors through loop closure detection and global optimization, and finally generate a high-precision map.
[0068] In summary, the high-precision map construction method provided in this application divides the point cloud set into three categories: ground points, corner points, and area points, and then creates the high-precision map. This not only increases the diversity of features, but also increases the constraint of the z-axis by processing the ground points separately, thus alleviating the problem of excessively rapid z-axis divergence.
[0069] SLAM technology includes not only front-end odometry but also back-end optimization. Back-end optimization often eliminates accumulated errors by adding loop closures. Loop closures typically employ matching methods such as ICP (Interlocking Proof-of-Care). However, if the accumulated error between two frames of LiDAR data is too large, it becomes difficult to calculate the correct loop closure pose. Therefore, global matching methods can be used for pose estimation, such as Cartographer, which uses CSM (Correlation Scan Match) for pose estimation. However, this significantly increases the time consumption for 6-DOF estimation, reducing the overall computational efficiency of mapping. To address the problem of excessive resource consumption in loop closure matching, this embodiment provides a loop closure detection method. For the specific process, please refer to [link to documentation]. Figure 2 :
[0070] Step 201: Perform loop closure detection on the point cloud set of each frame of LiDAR data.
[0071] Computer devices can perform loop closure detection on point cloud sets in many ways, and this embodiment does not limit the methods used.
[0072] If a loopback is detected, the computer device will obtain a set of point clouds from two frames of LiDAR data. The next step is to calculate the transformation between these two point cloud sets.
[0073] Step 202: If a loop is detected between the m-th point cloud set and the n-th point cloud set, then the ground points in the m-th point cloud set and the n-th point cloud set are matched to obtain the transformation values of the z-axis, roll angle and pitch angle between the LiDAR data of the m-th frame and the n-th frame.
[0074] Assuming the two detected LiDAR frames are the m-th and n-th frames, they correspond to the m-th and n-th point cloud sets. A computer can fix one point cloud set as the target point cloud set and calculate the transformation of the other point cloud set relative to the target point cloud set. The calculation process consists of two parts: the first part matches ground points in the m-th and n-th point cloud sets, and the second part matches corner points and face points in the m-th and n-th point cloud sets.
[0075] In solving the first part, the computer equipment uses the Normal-ICP method to process ground points, prioritizing the estimation of transformation values for the three degrees of freedom: [z, roll, pitch]. Here, z represents the z-axis, roll represents the roll angle, and pitch represents the pitch angle.
[0076] Step 203: Using the transformed values of the z-axis, roll angle, and pitch angle as initial values, calculate the transformed values of the x-axis, y-axis, and yaw angle between the LiDAR data of the m-th frame and the n-th frame.
[0077] Specifically, the computer equipment can use the transformation values of the z-axis, roll angle, and pitch angle as initial values to compress the corner points and surface points in the m-th and n-th point cloud sets into a two-dimensional map; and use the CSM scan matching algorithm to match the two-dimensional map to obtain the transformation values of the x-axis, y-axis, and yaw angle between the m-th and n-th frame lidar data.
[0078] The computer equipment can use the transformed values of the z-axis, roll angle, and pitch angle as initial values and then further process the corner points and polygon points. After compressing the corner points and polygon points into a two-dimensional map, the computer equipment can use the CSM method to match and calculate the transformed values of the three degrees of freedom [x, y, yaw]. Here, x represents the x-axis, y represents the y-axis, and yaw represents the yaw angle.
[0079] Step 204: Using the transformation values of the x-axis, y-axis, z-axis, yaw angle, roll angle, and pitch angle as initial values, the ICP algorithm is used to calculate the final transformation value between the LiDAR data of the m-th frame and the n-th frame.
[0080] The computer combines the transformation values obtained from the two calculations to obtain the transformation values for the six degrees of freedom [x, y, z, roll, pitch, yaw]. Due to discretization errors during the calculation process, the computer needs to use these six transformation values as initial values to perform an ICP matching on the m-th and n-th frames of LiDAR data to obtain the final, accurate transformation between the two frames of LiDAR data. This loop-based matching method improves the matching success rate and accuracy even with large accumulated errors, and also significantly reduces computation time, thus increasing efficiency.
[0081] If loop closure matching is introduced, step 104 can be replaced with: creating a high-precision map based on the pose and the final transformation value.
[0082] In this embodiment, by introducing ground points into the loop matching process, the estimation of the 6-DOF pose can be decoupled, achieving a higher matching success rate and accuracy under large cumulative errors, and shortening the computation time.
[0083] Please refer to Figure 3 This illustration shows a structural block diagram of a high-precision map building apparatus according to an embodiment of this application. This high-precision map building apparatus can be applied to a computer device. The high-precision map building apparatus may include:
[0084] The acquisition module 310 is used to acquire the point cloud set of the current frame and the previous frame of LiDAR data. The point cloud set includes ground points, corner points and surface points.
[0085] The acquisition module 310 is also used to acquire a point cloud distance model and calculate the distance between the same type of point cloud in the current frame and the previous frame based on the point cloud distance model.
[0086] The calculation module 320 is used to create an error function based on the distance and calculate the pose of the current frame relative to the previous frame based on the error function.
[0087] Create module 330 to create a high-precision map based on the pose.
[0088] In an optional embodiment, the calculation module 320 is further configured to:
[0089] When the point cloud distance model is the ground point distance model, the planar distance from each ground point in the point cloud set of the current frame to the ground in the previous frame is calculated, and the ground is composed of the ground points in the point cloud set of the previous frame.
[0090] In an optional embodiment, the calculation module 320 is further configured to:
[0091] When the point cloud distance model is a corner distance model, calculate the corner distance from each corner point in the point cloud set of the current frame to the corner point of the previous frame.
[0092] In an optional embodiment, the calculation module 320 is further configured to:
[0093] When the point cloud distance model is a face distance model, the planar distance from each face in the point cloud set of the current frame to the plane in the previous frame is calculated. The plane is composed of face points in the point cloud set of the previous frame.
[0094] Please refer to Figure 4 In an optional embodiment, the device further includes:
[0095] The detection module 340 is used to perform loop closure detection on the point cloud set of each frame of LiDAR data;
[0096] The matching module 350 is used to match the ground points in the m-th point cloud set and the n-th point cloud set if a loop is detected between the m-th point cloud set and the n-th point cloud set, so as to obtain the transformation values of the z-axis, roll angle and pitch angle between the m-th frame and the n-th frame of lidar data.
[0097] The calculation module 320 is also used to calculate the transformation values of the x-axis, y-axis and yaw angle between the LiDAR data of the m-th frame and the n-th frame, using the transformation values of the z-axis, roll angle and pitch angle as initial values;
[0098] The calculation module 320 is also used to take the transformation values of the x-axis, y-axis, z-axis, yaw angle, roll angle and pitch angle as initial values, and use the ICP algorithm to calculate the final transformation value between the lidar data of the m-th frame and the n-th frame.
[0099] In an optional embodiment, the calculation module 320 is further configured to:
[0100] Using the transformed values of the z-axis, roll angle, and pitch angle as initial values, the corner points and face points in the m-th point cloud set and the n-th point cloud set are compressed into a two-dimensional map;
[0101] The CSM scanning matching algorithm is used to match the two-dimensional map to obtain the transformation values of the x-axis, y-axis and yaw angle between the LiDAR data of the m-th frame and the n-th frame.
[0102] In an optional embodiment, module 330 is further configured to:
[0103] Create a high-precision map based on the pose and the final transformation value.
[0104] In summary, the high-precision map construction apparatus provided in this application divides the point cloud set into three categories: ground points, corner points, and area points before creating the high-precision map. This not only increases the diversity of features but also increases the constraint on the z-axis by processing the ground points separately, thus alleviating the problem of excessively rapid z-axis divergence.
[0105] By incorporating ground points into the loop matching process, the estimation of the 6-DOF pose can be decoupled, achieving a higher matching success rate and accuracy even with larger cumulative errors, while also shortening the computation time.
[0106] One embodiment of this application provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the high-precision map construction method described above.
[0107] One embodiment of this application provides a computer device including a processor and a memory, wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the high-precision map construction method described above.
[0108] It should be noted that the high-precision map building apparatus provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the high-precision map building apparatus can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the high-precision map building apparatus and the high-precision map building method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, which will not be repeated here.
[0109] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.
[0110] The above description is not intended to limit the embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of the embodiments of this application.
Claims
1. A method for constructing a high-precision map, characterized in that, The method includes: Obtain the point cloud set of the current frame and the previous frame of LiDAR data, wherein the point cloud set includes ground points, corner points and surface points; Obtain a point cloud distance model, and calculate the distance between the current frame and the same type of point cloud in the previous frame based on the point cloud distance model; An error function is created based on the distance, and the pose of the current frame relative to the previous frame is calculated based on the error function. Perform loop closure detection on the point cloud set of each frame of LiDAR data; If a loop is detected between the m-th point cloud set and the n-th point cloud set, then the ground points in the m-th point cloud set and the n-th point cloud set are matched to obtain the transformation values of the z-axis, roll angle and pitch angle between the lidar data of the m-th frame and the n-th frame. Using the transformed values of the z-axis, roll angle, and pitch angle as initial values, the corner points and face points in the m-th point cloud set and the n-th point cloud set are compressed into a two-dimensional map; The CSM scanning matching algorithm is used to match the two-dimensional map to obtain the transformation values of the x-axis, y-axis and yaw angle between the lidar data of the m-th frame and the n-th frame; Using the transformed values of the x-axis, y-axis, z-axis, yaw angle, roll angle, and pitch angle as initial values, the ICP algorithm is used to calculate the final transformed value between the LiDAR data of the m-th frame and the n-th frame. A high-precision map is created based on the pose and the final transformation value.
2. The method for constructing a high-precision map according to claim 1, characterized in that, The acquisition of the point cloud distance model, and the calculation of the distance between the current frame and the previous frame of the same type of point cloud based on the point cloud distance model, includes: When the point cloud distance model is a ground point distance model, the planar distance from each ground point in the point cloud set of the current frame to the ground in the previous frame is calculated, and the ground is composed of ground points in the point cloud set of the previous frame.
3. The method for constructing a high-precision map according to claim 1, characterized in that, The acquisition of the point cloud distance model, and the calculation of the distance between the current frame and the previous frame of the same type of point cloud based on the point cloud distance model, includes: When the point cloud distance model is a corner distance model, the corner distance from each corner point in the point cloud set of the current frame to the corner point of the previous frame is calculated.
4. The method for constructing a high-precision map according to claim 1, characterized in that, The acquisition of the point cloud distance model, and the calculation of the distance between the current frame and the previous frame of the same type of point cloud based on the point cloud distance model, includes: When the point cloud distance model is a face distance model, the planar distance from each face in the point cloud set of the current frame to the plane in the previous frame is calculated, and the plane is composed of face points in the point cloud set of the previous frame.
5. A high-precision map construction apparatus, characterized in that, The device includes: The acquisition module is used to acquire the point cloud set of the current frame and the previous frame of LiDAR data, wherein the point cloud set includes ground points, corner points and surface points; The acquisition module is further configured to acquire a point cloud distance model and calculate the distance between the current frame and the same type of point cloud in the previous frame based on the point cloud distance model. The calculation module is used to create an error function based on the distance and calculate the pose of the current frame relative to the previous frame based on the error function. The detection module is used to perform loop closure detection on the point cloud set of each frame of LiDAR data; The matching module is used to match the ground points in the m-th point cloud set and the n-th point cloud set if a loop is detected between them, so as to obtain the transformation values of the z-axis, roll angle and pitch angle between the lidar data of the m-th frame and the n-th frame. The calculation module is also used to compress the corner points and face points in the m-th point cloud set and the n-th point cloud set into a two-dimensional map by taking the transformation values of the z-axis, roll angle and pitch angle as initial values. The calculation module is also used to match the two-dimensional map using the CSM scanning matching algorithm to obtain the transformation values of the x-axis, y-axis and yaw angle between the lidar data of the m-th frame and the n-th frame; The calculation module is also used to take the transformation values of the x-axis, y-axis, z-axis, yaw angle, roll angle and pitch angle as initial values, and use the ICP algorithm to calculate the final transformation value between the m-th frame and the n-th frame of lidar data; A module is created to generate a high-precision map based on the pose and the final transformation value.
6. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, which is loaded and executed by a processor to implement the high-precision map construction method as described in any one of claims 1 to 4.
7. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing at least one instruction, which is loaded and executed by the processor to implement the method for constructing a high-precision map as described in any one of claims 1 to 4.