Point cloud map lightweight and matching auxiliary positioning method, system, device and medium

Abstract

The present invention relates to the field of navigation and positioning technology. The present invention provides a point cloud map lightweight and matching auxiliary positioning method, system, device and medium, which can obtain robust and high-precision registration results. The present invention uses neighborhood geometric relationships and statistical analysis methods to adaptively calculate the features of each point, designs a feature component processing algorithm to represent the point feature intensity, extracts feature salient points and reconstructs a lightweight map. In view of the problem that the matching iteration process is prone to fall into the local optimum, a coarse-fine matching scheme is designed, and the error function is constructed symmetric, the matching algorithm is improved, and the laser radar feature point cloud with motion distortion removed is registered with the constructed lightweight map to achieve accurate positioning in complex scenes.

Description

Technical Field

[0001] The present invention relates to the field of navigation and positioning technology, and in particular to a point cloud map lightweight and matching auxiliary positioning method, system, equipment and medium. Background Art

[0002] In recent years, driverless vehicles have received extensive attention in academia and industry due to their advantages of freeing up manpower and alleviating traffic pressure and the possibility of accidents. Obviously, driverless vehicles must accurately locate their location in order to effectively ensure the safe and reliable implementation of tasks such as intelligent path planning and automatic motion control. However, in environments with obstructions such as garages and high-rise buildings, GNSS is difficult to provide continuous and effective positioning information, and the existing inertial navigation / wheel speed / lidar mileage calculation methods cannot avoid the problem of error accumulation. High-precision map data is not affected by weather conditions, environment, light and other factors, and has the characteristics of high stability and high precision. Therefore, high-precision maps are indispensable for the positioning of unmanned vehicles in complex scenarios. However, in most scenarios, the amount of high-precision point cloud map data is large, and the computing resources of the on-board computing unit are limited, making it difficult to process a complete high-precision point cloud map. Lightweight processing of high-precision maps has become a necessary choice. At the same time, the lightweight process is bound to lead to the loss of environmental feature information. How to design a suitable lightweight and point cloud matching solution is a problem to be solved. Summary of the invention

[0003] In view of this, the present invention provides a point cloud map lightweight and matching auxiliary positioning method, system, device and medium, which can obtain robust and high-precision registration results.

[0004] In order to solve the above problems, the technical solution of the present invention is as follows:

[0005] A method for lightweight and matching auxiliary positioning of point cloud maps, which calculates the triplet features of set point pairs in the point cloud map to form a multi-dimensional feature vector of the center point, calculates the point feature intensity to filter out the feature-significant points, and forms a lightweight point cloud map; matches the laser radar scanning point cloud with motion distortion removed with the lightweight map, extracts features from the scanning point cloud, performs initial rough registration, and performs secondary fine registration of the registered point cloud with the lightweight map rough registration, and outputs the matching positioning result;

[0006] Among them, the point cloud map lightweight algorithm based on APFH features extracts the point cloud of the feature-significant area and reconstructs it into a lightweight map.

[0007] Among them, the APFH feature is used for matching in the coarse registration stage: randomly sample the scanned point cloud and lightweight map, perform nearest neighbor query on the sampling points in the map and point cloud frame based on the APFH feature descriptor, use the point pair distance error to construct the optimization objective function, perform nonlinear solution, calculate the corresponding error at this time, and repeat the sampling and matching process until the conditions are met, and take the rotation and displacement corresponding to the minimum error as the final result.

[0008] Among them, the matching and positioning algorithm based on the improved SAC-IA-SGICP is used to match the preprocessed lidar scanning point cloud with the lightweight point cloud map to obtain a globally consistent positioning result; among them, the APFH features are extracted from the lidar scanning point cloud with motion distortion removed, the SAC-IA algorithm is used to match the single-frame point cloud and the lightweight point cloud map, the error function is constructed symmetrically, and the Levenberg-Marquardt algorithm is used to iteratively solve the optimal rotation parameters to obtain the translation parameters.

[0009] Among them, IMU data is used to correct motion distortion: by retrieving IMU pre-integration, the relative pose increment of the start and end time of the point cloud of a specific frame is obtained, and the motion distortion of the point cloud frame is eliminated through coordinate transformation.

[0010] The present invention also provides a point cloud map lightweight and matching auxiliary positioning system, including a point cloud map lightweight module and a matching positioning module, the specific contents of which are as follows: the point cloud map lightweight module calculates the triplet features of the set point pairs in the point cloud map, constitutes a multi-dimensional feature vector of the center point, calculates the point feature intensity to screen the feature-significant points, and forms a lightweight point cloud map; the matching positioning module matches the laser radar scanning point cloud with motion distortion removed with the lightweight map, extracts features from the scanning point cloud, performs initial coarse alignment, and then performs secondary fine alignment of the registered point cloud with the lightweight map.

[0011] The present invention also provides an electronic device, which includes a processor and a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the point cloud map lightweight and matching assisted positioning method described in the present invention.

[0012] The present invention also provides a computer-readable storage medium, which stores a computer program, and the computer program is used to execute the point cloud map lightweight and matching assisted positioning method described in the present invention.

[0013] Beneficial effects:

[0014] 1. The method of the present invention uses the neighborhood geometric relationship and statistical analysis method to adaptively calculate the features of each point, designs a feature component processing algorithm to represent the point feature strength, extracts the feature-significant points to reconstruct the lightweight map. A coarse-fine matching scheme is designed to address the problem that the matching iteration process is prone to fall into the local optimum, and the error function is constructed symmetric, the matching algorithm is improved, and the laser radar feature point cloud with motion distortion removed is aligned with the constructed lightweight map to achieve accurate positioning in complex scenarios. Based on the APFH feature and applied to the point cloud map lightweight algorithm, the APFH feature considers the distribution of the geometric features of the point and the area around it at the same time, has regional adaptive capabilities for the geometric feature representation, and can significantly reduce the weight of abnormal point clouds. The feature component processing algorithm is designed according to the relationship between the distribution characteristics of the point feature components and the significance of the point features, and the compression rate and information entropy of the reconstructed point cloud map are higher.

[0015] 2. The lightweight method of the present invention considers the distribution of the geometric features of a single point and its regional geometric features in the vicinity at the same time, calculates the APFH (Adaptive Point Feature Histograms) features of each point using the neighborhood geometric relationship and statistical analysis method, and designs a feature component processing algorithm to represent the point feature intensity, thereby extracting feature-significant points to reconstruct a lightweight map. The lightweight method can adaptively change the geometric feature representation of a single point according to the regional feature distribution, thereby ensuring feature richness while improving the map compression rate. The matching and positioning process uses the SAC-IA-SGICP algorithm to match the laser radar scanning point cloud with motion distortion removed with the lightweight point cloud map to obtain the registration pose. .

[0016] 3. The present invention adopts an improved SAC-IA (Sample Consensus Initial Alignment) algorithm and uses APFH features for matching, which is consistent with the features extracted by the lightweight process, has better local-global matching effect, and provides reliable initial values ​​for precise matching, reducing the probability of falling into local optimality.

[0017] 4. The precise matching process of the present invention improves the GICP algorithm and designs the SGICP (Symmetric Generalized Iterative Closest Point) algorithm, which further improves the matching positioning accuracy by symmetrically constructing the error function, and finally obtains a robust and high-precision registration result.

[0018] 5. The system of the present invention is used to implement the method of the present invention, using the neighborhood geometric relationship and statistical analysis method to adaptively calculate the features of each point, designing a feature component processing algorithm to represent the feature strength of the point, extracting the feature salient points to reconstruct the lightweight map. In view of the problem that the matching iteration process is prone to fall into the local optimum, a coarse-fine matching scheme is designed, and the error function is constructed symmetric, the matching algorithm is improved, and the laser radar feature point cloud with motion distortion removed is registered with the constructed lightweight map to achieve accurate positioning in complex scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Schematic diagram of a point pair triplet feature calculation model in an embodiment of the present invention.

[0020] Figure 2 Schematic diagram of point pairs required for APFH features of target point q according to an embodiment of the present invention.

[0021] Figure 3 Schematic diagram of APFH feature comparison in different regions in an embodiment of the present invention.

[0022] Figure 4 The figure is a schematic diagram of the matching and positioning process used in the embodiment of the present invention.

[0023] Figure 5 This is a schematic diagram of the implementation principle of the system of the present invention.

[0024] Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0026] The present invention provides a point cloud map lightweight and matching auxiliary positioning method, comprising the following steps:

[0027] The triplet features of the set point pairs in the point cloud map are calculated to form a multi-dimensional feature vector of the center point. The point feature intensity is calculated to screen the feature-significant points to form a lightweight point cloud map. The laser radar scanning point cloud with motion distortion removed is matched with the lightweight map, and the features are extracted from the scanning point cloud. The improved SAC-IA algorithm is used for initial coarse alignment. After alignment, the point cloud is then finely aligned with the lightweight map for a second time, and the improved SGICP algorithm is used to output the matching positioning results.

[0028] Among them, the point cloud map lightweight algorithm based on the APFH feature extracts the point cloud of the feature-significant area and reconstructs it into a lightweight map. The most intuitive method is to calculate the corner points with significant curvature changes in the point cloud or the extreme points based on certain geometric features as feature points. These points usually carry significant geometric information and are consistent with human visual perception. However, since the distribution of points inside the point cloud is difficult to ensure to be smooth and continuous, some abnormal points inevitably exist in the point cloud and destroy the curvature changes of nearby point clouds. In order to solve this problem, the present invention proposes the APFH feature, which draws on the PFH feature element calculation method and statistical analysis method. When judging whether a point is a feature point, the geometric features of the point and the distribution of the geometric features of the area nearby are considered at the same time, which significantly reduces the abnormal weight, so that the geometric feature description of the point or the area has the ability to adapt to the distribution of regional features. The collected high-precision point cloud map is processed, and the triple features between the set point pairs are calculated to form a multi-dimensional feature vector of the center point. Figure 1 As shown, the three angles α, β, and γ are used as the triplet features of point Ps about point Pt, and the calculation formula is as follows:

[0029] α=(Pt-Ps)×Ns·Nt (1)

[0030]

[0031]

[0032] Where Ns is the Ps normal and Nt is the Pt normal.

[0033] Note that point q in the point cloud map is the center point, the 5 nearest neighbors of the center point are neighborhood points, and the non-center points and non-neighborhood points in the neighborhood of the 5 neighboring points are secondary neighborhood points, such as Figure 2 As shown in the figure, the triplet features of the center point about the neighborhood points, the triplet features of the neighborhood points about the secondary neighborhood points, and the triplet features of the center point about the secondary neighborhood points are calculated to form a multi-dimensional feature vector of the center point. If there are n feature elements of the secondary neighborhood points that are 0, the current secondary neighborhood point contributes less to the regional features, and the r value is increased to expand the feature expression range. The transformation formula of r with n is as follows:

[0034]

[0035] Among them, ε1 is the lower bound of r, the upper bound of r is ε2+ε1, and n0 is the set threshold of the number of zero feature elements.

[0036] The improved regional feature calculation criterion increases the feature relationship between the center point and the distant neighboring points, and adaptively transforms with the region, making the feature expression more flexible and rich.

[0037] Comparison of APFH features of different regions in the embodiment of the present invention Figure 3 As shown in the figure, if the geometric features between the point to be sought and its neighborhood point set are not obvious and the plane range is large, there must be a phenomenon that a single component is prominent and some components are close to 0; on the contrary, the more rugged the area where the point to be sought and its neighborhood point set is located and the more obvious the geometric features are, the more balanced the values ​​of each component are. Based on the above rules, the feature strength of each point is defined as follows:

[0038]

[0039] Where p is the mean of the characteristic component, p i As the feature component, feature points whose feature strength exceeds the threshold are selected to reconstruct into a lightweight point cloud map.

[0040] Furthermore, there is a problem of motion distortion in the laser radar scanning point cloud. The present invention uses IMU data to correct motion distortion. IMU provides acceleration and angular velocity information, and after pre-integration, continuous velocity, position, and attitude information can be obtained:

[0041]

[0042]

[0043]

[0044] where ω t and a t is the original measurement data of IMU at time t, are angular velocity and acceleration bias respectively, are the influence of angular velocity and acceleration white noise, R t is the rotation matrix from the world coordinate system to the carrier coordinate system, and g is the gravity vector.

[0045] By retrieving the IMU pre-integration, the relative pose increment of the start and end time of a specific frame point cloud can be obtained, and the motion distortion of the point cloud frame can be eliminated through coordinate transformation.

[0046] Among them, based on the improved SAC-IA-SGICP matching positioning algorithm, the pre-processed laser radar scanning point cloud is matched with the lightweight point cloud map to obtain the global consistency positioning result. Combined with the characteristics of the lightweight point cloud map, the present invention designs a two-step matching method. The process is as follows: Figure 4As shown in the figure. Since the point cloud map is lightweight based on the APFH feature, the APFH feature is used for matching in the coarse registration stage, which can make full use of the features retained in the lightweight map to achieve better matching results. APFH features are extracted from the laser radar scanning point cloud with motion distortion removed, and the SAC-IA algorithm is used to match the single-frame point cloud and the lightweight point cloud map. The scanned point cloud and lightweight map are randomly sampled, and the nearest neighbor query is performed on the sampling points in the map and point cloud frame based on the APFH feature descriptor. The optimization objective function is constructed using the point pair distance error for nonlinear solution.

[0047] Define the i-th point pair a i 、b i The centroid coordinates of n is the number of point clouds:

[0048]

[0049] R is the rotation matrix to be solved, and the point distance error can be expressed as:

[0050]

[0051] The Huber penalty function with good robustness is used to construct a nonlinear optimization problem:

[0052]

[0053] The Huber penalty function formula is as follows, where M is the set threshold:

[0054]

[0055] Use the Levenberg-Marquardt method to solve, obtain the optimal solution of the rotation matrix R, and then solve the translation matrix:

[0056]

[0057] The corresponding error at this time is calculated, and the sampling and matching process is repeated until the conditions are met, and the rotation and displacement corresponding to the minimum error are taken as the final result.

[0058] The SAC-IA algorithm is more suitable for local and global registration, but the registration accuracy is slightly poor. Therefore, the rough registration result is used as the initial value, and the SGICP algorithm is used for fine registration. The ICP algorithm is a common type of point cloud registration method. Its matching accuracy usually depends on the quality of the initial value and is prone to fall into the local optimum. The initial value provided by the SAC-IA algorithm can effectively reduce the probability of the GICP algorithm falling into the local optimum. The GICP algorithm integrates point-to-point, point-to-surface, and surface-to-surface registration, which effectively improves the registration accuracy. At the same time, the present invention improves the GICP algorithm, symmetricizes the construction error function, and designs the SGICP algorithm to further improve the registration accuracy.

[0059] Assuming that each point in the point cloud obeys Gaussian distribution,

[0060] The point covariance matrix is:

[0061]

[0062] in, is the k-neighborhood point of point A, is the neighborhood point mean, and similarly we can get

[0063] Symmetric construction error:

[0064]

[0065] In the case of the optimal solution,

[0066]

[0067] According to the maximum likelihood estimation idea, d i The probability of occurrence is the largest, and the optimization objective function is constructed.

[0068]

[0069] Where p(d i ) means d i Probability of occurrence, Available

[0070]

[0071] Since the determinant value of the rotation matrix in the three-dimensional rigid body transformation is 1,

[0072]

[0073] when hour, Degenerates into point-to-point registration;

[0074] when hour, It degenerates into point-to-plane registration, where P is the orthogonal projection matrix;

[0075] The optimal rotation parameters are iteratively solved using the Levenberg-Marquardt algorithm, and the translation parameters are obtained using formula (13). The result is the pose information of the solved LiDAR scan point cloud in the point cloud map.

[0076] In summary, the present invention considers the matching and positioning application after the map is lightweight, and designs a corresponding two-step matching scheme of coarse registration and fine registration. In the coarse registration stage, the APFH features extracted from the scanned point cloud are matched with the lightweight map by SAC-IA, making fuller use of the point cloud map information. The coarse registration result is used as the initial value to participate in the fine registration process, reducing the probability of fine registration falling into the local optimum, and proposing the SGICP algorithm to symmetric construct the error function, further improving the accuracy of the matching result.

[0077] The present invention also proposes a point cloud map lightweight and matching auxiliary positioning system, including a point cloud map lightweight module and a matching positioning module, the specific contents of which are as follows: the point cloud map lightweight module calculates the triplet features of the set point pairs in the point cloud map to form a multi-dimensional feature vector of the center point, calculates the point feature intensity to screen the feature-significant points, and forms a lightweight point cloud map; the matching positioning module matches the laser radar scanning point cloud with motion distortion removed with the lightweight map, extracts features from the scanning point cloud, performs initial coarse alignment, and then performs secondary fine alignment of the registered point cloud with the lightweight map.

[0078] The improved SAC-IA algorithm is used to implement the initial rough registration, and the improved SGICP algorithm is used to output the matching positioning result. The improved SGICP algorithm is the specific algorithm described in the method of the present invention. The system provided by the embodiment of the present invention can execute the point cloud map lightweight and matching auxiliary positioning method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. The implementation principle of the system of the present invention is as follows: Figure 5 shown.

[0079] A specific implementation step of a point cloud map lightweight and matching auxiliary positioning system of the present invention is as follows:

[0080] 1. Calculate the APFH features for each point in the high-precision point cloud map;

[0081] 2. Traverse the point cloud and calculate the feature strength of each point;

[0082] 3. Filter points with feature strength greater than 0.3 and reconstruct a lightweight point cloud map;

[0083] 4. Obtain IMU acceleration and angular velocity information and build an IMU pre-integration module;

[0084] 5. Receive the laser radar scanning point cloud and remove redundant points;

[0085] 6. Retrieve the IMU pre-integration module to obtain the relative position increment of the start and end time of a specific frame point cloud, and eliminate the motion distortion of the point cloud frame through coordinate transformation;

[0086] 7. Extract the APFH features of the scanned point cloud frame after removing motion distortion;

[0087] 8. Randomly sample the scan point cloud frame and the lightweight map, and match the sampling area with the highest similarity based on the APFH feature descriptor;

[0088] 9. Use the distance error of the closest point pair to construct the Huber penalty function, use the LM algorithm to solve the nonlinear problem, and obtain the rotation matrix and translation matrix;

[0089] 10. Calculate the current matching error. If it does not meet the threshold requirement, exclude the current matching solution and repeat steps 8-9. If it meets the threshold requirement, proceed to the next step.

[0090] 11. Use the matching result of the previous step to transform the scanned point cloud and record it as the source point cloud in subsequent operations;

[0091] 12. Symmetrically construct the error function and design nonlinear optimization problems;

[0092] 13. Use LM algorithm to iteratively solve nonlinear problems;

[0093] 14. Calculate the current matching error. If it does not meet the threshold requirement, use the current result as the initial value and repeat steps 11-13. If it meets the threshold requirement, the matching ends.

[0094] 15. The product of the transformation matrices obtained from the two registrations is the pose conversion result of the current scan frame matching the point cloud map.

[0095] The embodiment of the present application also provides an electronic device, Figure 6The structure of the electronic device provided by the embodiment of the present invention is shown. For example, the electronic device 60 may include a processor 61, a memory 62 and a transmission device 63. The processor 61 is used to execute the point cloud map lightweight and matching auxiliary positioning method mentioned in the above embodiment, wherein the processor and the memory may be connected by a bus or other means, taking the bus connection as an example. The transmission device may be connected to the processor and the memory by wire or wirelessly. The memory, as a non-transient computer-readable storage medium, may be used to store non-transient software programs, non-transient computer executable programs and modules, such as program instructions / modules corresponding to the point cloud map lightweight and matching auxiliary positioning method in the embodiment of the present application. The processor executes various functional applications and data processing of the processor by running the non-transient software programs, instructions and modules stored in the memory, that is, the point cloud map lightweight and matching auxiliary positioning method in the above method embodiment is realized. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The one or more modules are stored in the memory, and when executed by the processor, the point cloud map lightweight and matching auxiliary positioning method in the embodiment is executed.

[0096] As another aspect, the present application also provides a computer-readable storage medium, which may be a computer-readable storage medium included in the device described in the above embodiment; or it may be a computer-readable storage medium that exists independently and is not assembled into the device. The computer-readable storage medium may be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field. The computer-readable storage medium stores one or more programs, and the programs are used by one or more processors to execute the point cloud map lightweight and matching auxiliary positioning method described in the present application.

[0097] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A point cloud map lightweight and matching auxiliary positioning method, characterized in that: Calculate the triplet features of the set point pairs in the point cloud map to form a multi-dimensional feature vector of the center point, calculate the point feature strength to filter out the feature-significant points, and form a lightweight point cloud map; match the laser radar scanning point cloud with motion distortion removed with the lightweight map, extract features from the scanning point cloud, perform initial rough registration, and then perform secondary fine registration with the lightweight map rough registration after registration of the point cloud, and output the matching positioning result; Among them, the point cloud map lightweight algorithm based on APFH features extracts the point cloud of the feature-significant area and reconstructs it into a lightweight map; In the coarse registration stage, APFH features are used for matching: randomly sample the scanned point cloud and lightweight map, perform nearest neighbor queries on the sampling points in the map and point cloud frame based on the APFH feature descriptor, construct an optimization objective function using the point pair distance error, perform nonlinear solution, calculate the corresponding error at this time, and repeat the sampling and matching process until the conditions are met, and take the rotation and displacement corresponding to the minimum error as the final result; Based on the improved SAC-IA-SGICP matching positioning algorithm, the pre-processed LiDAR scanning point cloud is matched with the lightweight point cloud map to obtain a globally consistent positioning result. The APFH feature is extracted from the LiDAR scanning point cloud with motion distortion removed, and the SAC-IA algorithm is used to match the single-frame point cloud and the lightweight point cloud map. The error function is constructed symmetrically, and the Levenberg-Marquardt algorithm is used to iteratively solve the optimal rotation parameters and obtain the translation parameters. Define the i-th point pair a i , b i The centroid coordinates of n is the number of point clouds: R is the rotation matrix to be solved, and the point distance error can be expressed as: The Huber penalty function with good robustness is used to construct a nonlinear optimization problem: The Huber penalty function formula is as follows, where M is the set threshold: Use the Levenberg-Marquardt method to solve, obtain the optimal solution of the rotation matrix R, and then solve the translation matrix: Calculate the corresponding error at this time, and repeat the sampling and matching process until the conditions are met, and take the rotation and displacement corresponding to the minimum error as the final result; Assuming that each point in the point cloud obeys Gaussian distribution, The point covariance matrix is: in, is the k-neighborhood point of point A, is the neighborhood point mean, and similarly we can get Symmetric construction error: In the optimal solution case, According to the maximum likelihood estimation idea, d i The probability of occurrence is the largest, and the optimization objective function is constructed. Where p(d i ) means d i Probability of occurrence, Available Since the determinant value of the rotation matrix in the three-dimensional rigid body transformation is 1, when hour, Degenerates into point-to-point registration; when hour, It degenerates into point-to-plane registration, where P is the orthogonal projection matrix; The Levenberg-Marquardt algorithm is used to iteratively solve the optimal rotation parameters and obtain the translation parameters; the result is the pose information of the solved LiDAR scanning point cloud in the point cloud map.

2. The method according to claim 1, characterized in that Use IMU data to correct motion distortion: By retrieving IMU pre-integration, the relative position increment of the start and end time of a specific frame point cloud is obtained, and the motion distortion of the point cloud frame is eliminated through coordinate transformation.

3. A point cloud map lightweight and matching auxiliary positioning system, characterized in that: It includes a point cloud map lightweight module and a matching and positioning module. The specific contents are as follows: the point cloud map lightweight module calculates the triplet features of the set point pairs in the point cloud map to form a multi-dimensional feature vector of the center point, calculates the point feature intensity to filter out the feature-significant points, and forms a lightweight point cloud map; the matching and positioning module matches the laser radar scanning point cloud with motion distortion removed with the lightweight map, extracts features from the scanning point cloud, performs initial rough registration, and then performs secondary fine registration of the registered point cloud with the lightweight map; In the coarse registration stage, APFH features are used for matching: randomly sample the scanned point cloud and lightweight map, perform nearest neighbor queries on the sampling points in the map and point cloud frame based on the APFH feature descriptor, construct an optimization objective function using the point pair distance error, perform nonlinear solution, calculate the corresponding error at this time, and repeat the sampling and matching process until the conditions are met, and take the rotation and displacement corresponding to the minimum error as the final result; Based on the improved SAC-IA-SGICP matching positioning algorithm, the pre-processed LiDAR scanning point cloud is matched with the lightweight point cloud map to obtain a globally consistent positioning result. The APFH feature is extracted from the LiDAR scanning point cloud with motion distortion removed, and the SAC-IA algorithm is used to match the single-frame point cloud and the lightweight point cloud map. The error function is constructed symmetrically, and the Levenberg-Marquardt algorithm is used to iteratively solve the optimal rotation parameters and obtain the translation parameters. Define the i-th point pair a i , b i The centroid coordinates of n is the number of point clouds: R is the rotation matrix to be solved, and the point distance error can be expressed as: The Huber penalty function with good robustness is used to construct a nonlinear optimization problem: The Huber penalty function formula is as follows, where M is the set threshold: Use the Levenberg-Marquardt method to solve, obtain the optimal solution of the rotation matrix R, and then solve the translation matrix: Calculate the corresponding error at this time, and repeat the sampling and matching process until the conditions are met, and take the rotation and displacement corresponding to the minimum error as the final result; Assuming that each point in the point cloud obeys Gaussian distribution, The point covariance matrix is: in, is the k-neighborhood point of point A, is the neighborhood point mean, and similarly we can get Symmetric construction error: In the optimal solution case, According to the maximum likelihood estimation idea, d i The probability of occurrence is the largest, and the optimization objective function is constructed. Where p(d i ) means d i Probability of occurrence, Available Since the determinant value of the rotation matrix in the three-dimensional rigid body transformation is 1, when hour, Degenerates into point-to-point registration; when hour, It degenerates into point-to-plane registration, where P is the orthogonal projection matrix; The Levenberg-Marquardt algorithm is used to iteratively solve the optimal rotation parameters and obtain the translation parameters; the result is the pose information of the solved LiDAR scanning point cloud in the point cloud map.

4. An electronic device, characterized in that: The electronic device includes a processor and a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the point cloud map lightweight and matching assisted positioning method described in any one of claims 1-3 above.

5. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and the computer program is used to execute the point cloud map lightweight and matching assisted positioning method described in any one of claims 1 to 3.

Images