Three-dimensional mapping method, device and storage medium for unmanned mine

Abstract

The disclosure provides a three-dimensional mapping method, device and storage medium for unmanned mine, belonging to the technical field of mining. The method comprises: optimizing the initial global pose of the laser radar according to the point cloud features obtained by the laser radar to obtain a first global pose, splicing the point cloud according to the first global pose to obtain a first subgraph, and registering a second subgraph adjacent to the first subgraph, taking the pose transformation relationship between the first subgraph and the second subgraph as a laser odometry factor; optimizing the first global pose according to the error of at least one of the GPS prior factor and the closed loop detection factor and the laser odometry factor to obtain a second global pose; aligning multiple first subgraphs to a global coordinate system, and splicing the multiple first subgraphs into a three-dimensional map according to the second global pose. Through multi-factor optimization and multi-level optimization, a more accurate and comprehensive three-dimensional map can be constructed, and the problems of map degradation and error accumulation in the modeling process are reduced.

Description

TECHNICAL FIELD

[0001] The present disclosure belongs to the technical field of mining, and particularly relates to a three-dimensional mapping method, device and storage medium for an unmanned mine. BACKGROUND

[0002] With the rapid development of technologies such as unmanned driving, 5G communication, artificial intelligence and cloud computing, the unmanned transportation system of an open-pit mine has ushered in rapid development, and landing operation schemes have appeared one after another in major open-pit mines in China. The unmanned transportation system can effectively reduce labor costs, improve production efficiency and ensure the safety of mine personnel. The unmanned transportation system is composed of a wireless network communication system (4G, 5G, wireless Mesh, etc.), a fleet management system located in a control center, unmanned mine dump trucks, loading auxiliary operation equipment (excavators, electric shovels, etc.), unloading auxiliary operation equipment (bulldozers, loaders, etc.), road maintenance auxiliary equipment (graders, water trucks, etc.), and service vehicles (pickups, etc.).

[0003] In an open-pit mine, unmanned mine trucks combine standard maps and real-time positioning information to perform loading operations, which is becoming an important development direction for unmanned mines. In order to accurately control the unmanned mine trucks, accurate map information must be provided first. SUMMARY

[0004] One technical problem to be solved by the present disclosure is to provide a three-dimensional mapping method, device and storage medium for an unmanned mine, which can improve the accuracy of a three-dimensional map.

[0005] According to an aspect of the present disclosure, a three-dimensional mapping method for an unmanned mine is provided, which includes: optimizing an initial global pose of a laser radar according to point cloud features obtained by the laser radar located on a mine truck to obtain a first global pose, and splicing point clouds according to the first global pose to obtain a first sub-map; registering a second sub-map adjacent to the first sub-map according to the first global pose, and taking a pose transformation relationship between the first sub-map and the second sub-map as a laser odometry factor; optimizing the first global pose according to an error of at least one of a global positioning system (GPS) prior factor and a loop closure detection factor and the laser odometry factor to obtain a second global pose, wherein the GPS prior factor is displacement information in the initial global pose, and the loop closure detection factor is a pose transformation relationship between the first sub-map and a loop closure detection candidate sub-map corresponding to the first sub-map; and aligning a plurality of first sub-maps to a global coordinate system, and splicing the plurality of first sub-maps into a three-dimensional map according to the second global pose.

[0006] In some embodiments, an external parameter between the laser radar and a center of a rear axle of the vehicle body of the mine truck is obtained, and global positioning information of the center of the rear axle of the vehicle body is obtained by a global navigation satellite system (GNSS) integrated inertial navigation system located on the mine truck; and an initial global pose of the laser radar is obtained according to the external parameter and the global positioning information.

[0007] In some embodiments, the first subgraph is taken as a loop closure detection unit to perform loop closure detection, and a loop closure detection candidate subgraph corresponding to the first subgraph is determined.

[0008] In some embodiments, taking the first subgraph as a loop closure detection unit to perform loop closure detection, and determining a loop closure detection candidate subgraph corresponding to the first subgraph includes: obtaining a plurality of historical subgraphs corresponding to the first subgraph; sorting the plurality of historical subgraphs in chronological order to form a historical subgraph sequence; sequentially registering the first subgraph and each historical subgraph in the historical subgraph sequence by using the first global pose; splicing the first subgraph and the registered historical subgraph, and finding all point pairs within a first preset distance in the first subgraph and the registered historical subgraph, and calculating distance values between all the point pairs; if the distance value is less than a distance threshold, the registered historical subgraph is taken as the loop closure detection candidate subgraph corresponding to the first subgraph.

[0009] In some embodiments, obtaining the plurality of historical subgraphs corresponding to the first subgraph includes: constructing a screening area with the first frame laser radar coordinate system global position in the first subgraph as the center and a second preset distance as the radius, and determining the plurality of historical subgraphs in the screening area.

[0010] In some embodiments, obtaining the first global pose includes: calculating the relative rotation time of each point for each frame of point cloud obtained by the laser radar; calculating the motion increment of the laser radar based on the initial global pose; linearly interpolating the motion increment by using the relative rotation time to unify the positions of all points in each frame of point cloud to the time when the frame of point cloud starts; converting the processed points to the global coordinate system; performing gridding processing on the processed points, counting the number of points in each grid, and if the number of points in the grid is greater than a number threshold, calculating the mean and covariance of all points in the grid, performing eigenvalue decomposition on the covariance matrix to obtain a feature matrix containing eigenvalues; constructing a least squares constraint based on the feature matrix, wherein the motion increment is taken as the initial value of the least squares constraint, the least squares constraint is solved, and the first global pose is determined.

[0011] In some embodiments, constructing the least squares constraint based on the feature matrix includes: judging whether the point cloud feature in the grid is a surface feature, a line feature or a Gaussian distribution feature according to the eigenvalues of the feature matrix, and adjusting the eigenvalues in the inverse matrix corresponding to the feature matrix according to the judgment result; determining the result of the eigenvalue decomposition of the covariance matrix according to the inverse matrix corresponding to the adjusted feature matrix, and constructing the least squares constraint according to the result.

[0012] In some embodiments, judging the point cloud feature in the grid as a surface feature, a line feature or a Gaussian distribution feature according to the eigenvalues of the feature matrix, and adjusting the eigenvalues in the inverse matrix corresponding to the feature matrix according to the judging result comprises: in the case that the feature matrix is a diagonal matrix, the feature matrix comprises a first eigenvalue, a second eigenvalue and a third eigenvalue, wherein the first eigenvalue is greater than the second eigenvalue, and the second eigenvalue is greater than the third eigenvalue, if the difference between the first eigenvalue and the second eigenvalue is greater than a first threshold, it is determined that the point cloud feature in the grid is a line feature, and a fourth eigenvalue in the inverse matrix corresponding to the feature matrix is set to 0, wherein the fourth eigenvalue and the first eigenvalue are reciprocals of each other; if the difference between the first eigenvalue and the second eigenvalue is less than or equal to the first threshold, and the difference between the second eigenvalue and the third eigenvalue is greater than a second threshold, it is determined that the point cloud feature in the grid is a surface feature, and the fourth eigenvalue and a fifth eigenvalue in the inverse matrix corresponding to the feature matrix are set to 0, wherein the fifth eigenvalue and the second eigenvalue are reciprocals of each other; if the difference between the second eigenvalue and the third eigenvalue is less than or equal to the second threshold, it is determined that the point cloud feature in the grid is a Gaussian distribution feature, and the fourth eigenvalue, the fifth eigenvalue and a sixth eigenvalue in the inverse matrix corresponding to the feature matrix are not adjusted, wherein the sixth eigenvalue and the third eigenvalue are reciprocals of each other.

[0013] In some embodiments, splicing the point cloud according to the first global pose to obtain the first subgraph comprises: splicing the point cloud of the continuous N frames after distortion removal according to the first global pose to obtain the first subgraph.

[0014] In some embodiments, registering the second subgraph adjacent to the first subgraph according to the first global pose, and taking the pose transformation relationship between the first subgraph and the second subgraph as a laser odometry factor comprises: registering the first subgraph and the second subgraph in the first frame coordinate system of the first subgraph by using the first global pose to obtain the pose transformation relationship between the first subgraph and the second subgraph, and taking the pose transformation relationship as the laser odometry factor.

[0015] In some embodiments, the three-dimensional map in the global coordinate system is converted to the geographic coordinate system.

[0016] According to another aspect of the present disclosure, a three-dimensional mapping device for unmanned mine area is also provided, comprising: a stitching module configured to optimize an initial global pose of a laser radar according to point cloud features obtained by the laser radar located on a mine car, to obtain a first global pose, and to stitch the point cloud according to the first global pose to obtain a first subgraph; a laser odometry factor obtaining module configured to register a second subgraph adjacent to the first subgraph according to the first global pose, and to take a pose transformation relationship between the first subgraph and the second subgraph as a laser odometry factor; and a mapping module configured to optimize the first global pose according to an error of at least one of a global positioning system (GPS) prior factor and a loop closure detection factor and the laser odometry factor, to obtain a second global pose, wherein the GPS prior factor is displacement information in the initial global pose, the loop closure detection factor is a pose transformation relationship between the first subgraph and a loop closure detection candidate subgraph corresponding to the first subgraph, a plurality of first subgraphs are aligned to a global coordinate system, and the plurality of first subgraphs are stitched into a three-dimensional map according to the second global pose.

[0017] In some embodiments, the three-dimensional mapping device further comprises: an external parameter obtaining module configured to obtain an external parameter between the laser radar and a center of a rear axle of a vehicle body of the mine car, and to obtain global positioning information of the center of the rear axle of the vehicle body by using a global navigation satellite system (GNSS) integrated inertial navigation located on the mine car; and a GPS prior factor obtaining module configured to obtain the initial global pose of the laser radar according to the external parameter and the global positioning information, and to take displacement information in the initial global pose as the GPS prior factor.

[0018] In some embodiments, the three-dimensional mapping device further comprises: a loop closure detection factor obtaining module configured to perform loop closure detection by taking the first subgraph as a loop closure detection unit to determine a loop closure detection candidate subgraph corresponding to the first subgraph.

[0019] According to another aspect of the present disclosure, a three-dimensional mapping device for unmanned mine area is also provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute a three-dimensional mapping method as described above based on instructions stored in the memory.

[0020] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium having computer program instructions stored thereon is also provided, the instructions being executed by a processor to implement a three-dimensional mapping method as described above.

[0021] In the embodiments of the present disclosure, the global pose of the laser radar is optimized through errors of the GPS prior factor and the loop closure detection factor and the laser odometry factor, and then a plurality of subgraphs are stitched into a three-dimensional map based on the global pose, that is, a more accurate and comprehensive three-dimensional map can be constructed through multi-factor optimization and multi-level optimization, and the problems of map degradation and error accumulation in the three-dimensional map modeling process are reduced.

[0022] Other features of the present disclosure, its nature and advantages will become apparent from the detailed description of exemplary embodiments of the present disclosure which follows, taken with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0023] The accompanying drawings, which form a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0024] The present disclosure can be more clearly understood with reference to the following detailed description in connection with the attached drawings, wherein:

[0025] Figure 1 Flowchart of some embodiments of the method for three-dimensional mapping of an unmanned mine according to the present disclosure;

[0026] Figure 2 Installation of components in some embodiments according to the present disclosure;

[0027] Figure 3 Flowchart of some other embodiments of the method for three-dimensional mapping of an unmanned mine according to the present disclosure;

[0028] Figure 4 Layered optimization in some embodiments according to the present disclosure;

[0029] Figure 5 Structure diagram of some embodiments of the device for three-dimensional mapping of an unmanned mine according to the present disclosure;

[0030] Figure 6 Structure diagram of some other embodiments of the device for three-dimensional mapping of an unmanned mine according to the present disclosure; and

[0031] Figure 7 Structure diagram of some other embodiments of the device for three-dimensional mapping of an unmanned mine according to the present disclosure. DETAILED DESCRIPTION

[0032] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. Note that the relative arrangement, numerical expressions, and numerical values of components and steps set forth in these embodiments are not limiting to the scope of the present disclosure unless otherwise specifically stated.

[0033] It should be understood, however, that the actual dimensions of the parts shown in the accompanying drawings can not be drawn to scale.

[0034] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way limiting to the scope of the present disclosure and its applications or uses.

[0035] Techniques, methods, and devices known to those of ordinary skill in the relevant art(s) can not be discussed in detail herein. However, where appropriate, such techniques, methods, and devices can be considered part of the present disclosure.

[0036] In all of the examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not limiting. Thus, other examples of the example embodiments can have different values.

[0037] It should be noted that like reference numerals and letters refer to like items throughout the attached drawings, and once an item is defined in one drawing, it is not necessary to discuss it further in subsequent drawings.

[0038] For the purpose of making the purposes, technical solutions and advantages of the present disclosure more clear, the present disclosure is further described in detail below with reference to the specific embodiments and with reference to the drawings.

[0039] At present, there are two methods for modeling mine area maps, one is a two-dimensional map construction method, and the other is a three-dimensional map construction method.

[0040] The two-dimensional map construction method is to collect road edge information of a loading area in a mine area through a single GNSS (Global Navigation Satellite System) combined with an IMU (Inertial Measurement Unit), and then to construct a two-dimensional map containing only road edge information according to the position information. This two-dimensional mapping method relies on the accurate collection of all road edge information of the loading area by the mine card, and the process of collecting all road edge information is relatively cumbersome and time-consuming. Moreover, when the mine card drives to a narrow area, it often cannot collect road edge information, thereby failing to guarantee the integrity of the collected road edge information, and further making the constructed two-dimensional map less accurate. In specific practical applications, the two-dimensional map can only provide geographic information in one plane, and thus cannot effectively provide precise road surface information for the unmanned mine vehicle in some mine areas with large fluctuations. Therefore, the three-dimensional map comes into being.

[0041] Another mode adopted by the three-dimensional map construction method is a combined measurement mode, which combines an IMU and a laser radar to construct a three-dimensional map of a mine area. The scheme of using an IMU and a laser radar to fuse mapping has gradually become mainstream in recent years, and can work well in situations where laser radar mapping is prone to failure (bumpy ground, rapid motion), and the mapping accuracy has also been significantly improved. The scheme of using an IMU and a laser radar to fuse mapping is divided into two categories: a filtering-based scheme and an optimization-based scheme. The filtering-based scheme usually uses an IMU to propagate the state, and then updates the state according to the registration result of the laser radar point cloud. The optimization-based scheme models the mapping problem as a graph optimization problem, with optimization variables as the vertices of the graph and laser odometry and IMU pre-integration measurements as the edges of the graph. The optimization-based scheme can optimize historical measurements and current measurements together, while the filtering-based scheme generally only considers the measurement information of adjacent time points.

[0042] In the related art, since the loading area of a mine area is usually open and has no obvious geometric features, only the laser radar is used to scan the terrain to obtain point clouds, and then the point clouds are spliced to obtain a three-dimensional map. This method is prone to cause map degradation. By using the combined measurement mode to construct a three-dimensional map, whether it is a filtering-based scheme or an optimization-based scheme, historical observation information is used to infer current information, and only relative position information can be obtained, and absolute position information cannot be obtained. Therefore, when an error occurs in a position of measurement, the error will be accumulated during repeated measurement in the subsequent, and point cloud ghosting will occur, causing the problem of layered maps, thereby causing the accuracy of the map to decrease. In addition, the map constructed by this scheme cannot be aligned with the geographic coordinate system.

[0043] Figure 1 A flowchart of some embodiments of the three-dimensional mapping method of the mine area of the present disclosure.

[0044] In step 110, the initial global pose of the laser radar is optimized according to the point cloud features obtained by the laser radar located on the mine car, to obtain a first global pose, and the first global pose is used to splice the point cloud to obtain a first sub-map.

[0045] In some embodiments, as shown in Figure 2 The laser radar 04 is installed at a position in front of, left of, or right of the mine car, and the installation position of the laser radar should ensure that it has a sufficient field of view to ensure that it can smoothly scan the geographic environment. The laser radar itself coordinate system is set to, for example, the laser radar coordinate system is a right-handed system, the X axis is parallel to the mine car and points to the forward direction of the mine car, the Y axis points to the left side of the mine car, and the Z axis points upward.

[0046] In some embodiments, for each frame of point cloud acquired by the laser radar, a relative rotation time of each point is calculated; based on the initial global pose, a motion increment of the laser radar is calculated; all points in each frame of point cloud are unified to a time when the frame of point cloud starts by linearly interpolating the motion increment with the relative rotation time; the processed points are converted to a global coordinate system; the processed points are rasterized to count the number of points in each grid, and if the number of points in the grid is greater than a threshold, the mean and covariance of all points in the grid are calculated, the covariance matrix is eigenvalue decomposed to obtain a feature matrix containing eigenvalues; based on the feature matrix, a least squares constraint is constructed, wherein the motion increment is taken as an initial value of the least squares constraint, the least squares constraint is solved to determine the first global pose.

[0047] In some embodiments, the first global pose is used to align the first subgraph and the second subgraph in the first frame coordinate system of the first subgraph to obtain a pose transformation relationship between the first subgraph and the second subgraph, and the pose transformation relationship is taken as a laser odometry factor.

[0048] In step 120, a second subgraph adjacent to the first subgraph is registered according to the first global pose, and a pose transformation relationship between the first subgraph and the second subgraph is taken as a laser odometry factor.

[0049] In some embodiments, the first global pose is used to align the first subgraph and the second subgraph in the first frame coordinate system of the first subgraph to obtain a pose transformation relationship between the first subgraph and the second subgraph, and the pose transformation relationship is taken as a laser odometry factor.

[0050] In step 130, the first global pose is optimized according to an error of at least one of a GPS prior factor and a loop detection factor and the laser odometry factor to obtain a second global pose, wherein the GPS prior factor is displacement information in the initial global pose, and the loop detection factor is a pose transformation relationship between the first subgraph and a loop detection candidate subgraph corresponding to the first subgraph.

[0051] In step 140, a plurality of first subgraphs are aligned to a global coordinate system, and the plurality of first subgraphs are spliced into a three-dimensional map according to the second global pose.

[0052] In the above embodiments, the global pose of the laser radar is optimized through the error of the GPS prior factor and the loop detection factor and the laser odometry factor, and then a plurality of subgraphs are spliced into a three-dimensional map based on the global pose, that is, through multi-factor optimization and multi-level optimization, a more accurate and comprehensive three-dimensional map can be constructed, and the problems of map degradation and error accumulation in the three-dimensional map modeling process are reduced.

[0053] Figure 3 Flowchart of another embodiment of the three-dimensional mapping method of the unmanned mine area of the present disclosure.

[0054] In step 310, the external parameters between the laser radar and the center of the rear axle of the mine truck are obtained, and the global positioning information of the center of the rear axle of the mine truck is obtained by using the GNSS integrated inertial navigation system located on the mine truck.

[0055] In some embodiments, a coordinate system of the mine truck is set, for example, with the center of the rear axle of the mine truck as the origin, the X axis of the center of the rear axle of the mine truck pointing to the forward direction of the mine truck, the Y axis pointing to the right side of the mine truck, and the Z axis pointing downward, and the laser radar and the GNSS integrated inertial navigation system are installed on the mine truck. Figure 2 As shown in the figure, the main and auxiliary antennas of the GNSS integrated inertial navigation system are installed at the front end of the mine truck, with the main antenna 01 on the left and the auxiliary antenna 02 on the right. The auxiliary antenna 02 points to the main antenna and is perpendicular to the vehicle body. It is necessary to ensure that the GPS antenna is not lower than the rear hopper, otherwise it will affect the reception of the GPS signal. The GNSS integrated inertial navigation system is installed on the center axis 03 of the vehicle body.

[0056] In some embodiments, obtaining the global positioning information of the center of the rear axle of the mine truck by using the GNSS integrated inertial navigation system located on the mine truck comprises: manually measuring the bias parameters of the GNSS integrated inertial navigation system relative to the center of the rear axle of the mine truck, inputting the bias parameters into the GNSS integrated inertial navigation system host computer software, and obtaining the global pose of the center of the rear axle of the mine truck, wherein the coordinate system of the center of the rear axle of the mine truck is the same as the coordinate system of the mine truck.

[0057] In some embodiments, a scene with certain geometric constraints is selected in the mine area, the mine truck is made to drive at a uniform speed around a predetermined path, laser radar and GNSS integrated inertial navigation system data are collected, the external parameters between the laser radar and the center of the rear axle of the mine truck are calibrated, and the pose of the laser radar in the coordinate system of the center of the rear axle of the mine truck is estimated.

[0058] For example, the mine truck is made to drive around an 8-shaped path in the loading area, and laser radar point cloud and GNSS integrated inertial navigation system positioning output are collected for a predetermined time (e.g. more than 3 minutes). The position of the laser radar relative to the center of the rear axle of the mine truck is measured by a tape measure According to the above installation and configuration of the laser radar and the GNSS integrated inertial navigation system, it can be known that the initial external parameters between the laser radar coordinate system and the center of the rear axle of the mine truck coordinate system are The relative pose of the laser radar in the first frame of the laser radar coordinate system at continuous time is obtained. For example, the least squares constraint is applied to obtain the relative pose of the laser radar in the first frame of the laser radar coordinate system at continuous time. The global pose of the center of the rear axle of the mine truck output by the GNSS integrated inertial navigation system is transformed into a relative pose, combined with the above relative pose of the current laser radar in the first frame of the laser radar coordinate system, and the external parameters between the laser radar and the center of the rear axle of the mine truck are estimated by using the eye-hand calibration method, i.e. the pose of the laser radar in the coordinate system of the center of the rear axle of the mine truck is estimated. The optimization initial value of the external parameters is

[0059] Since the loading area is relatively flat, the mining truck can be considered to be moving in a plane. Therefore, the Z-axis translation of the external parameter may not be estimated accurately. If the final estimated value differs significantly from the measured value, the measured value shall prevail.

[0060] In step 320, the initial global pose of the lidar is obtained based on the external parameters and the global positioning information, and the displacement information in the initial global pose is used as a GPS prior factor.

[0061] In some embodiments, the point cloud data collected by the laser radar at time t is obtained, a time threshold τ (τ>0) is set, and the two closest timestamps t in the time period t-τ to t+τ are found by the nearest neighbor search method. l , t r (t r >t l ), the corresponding GNSS combined inertial navigation positioning data are right Perform linear interpolation to calculate the global positioning information of the rear axle center of the vehicle at time t Where a=(t r -t) / (t r -t l ), b=(tt l ) / (t r -t l ). The initial global pose of the lidar in, is the position of the rear axle center of the vehicle in the global coordinate system at time t, It is an external reference. The displacement part of is added to the factor graph as a GPS prior factor.

[0062] In step 330 , the initial global pose of the laser radar is optimized based on the point cloud features acquired by the laser radar located on the mining car to obtain a first global pose.

[0063] In some embodiments, the point p of each frame of point cloud acquired by the laser radar is i =(xyz) T Calculate its relative rotation angle θ i = atan(y / x), and then calculate the relative rotation time t of each point i . Calculate the motion increment of the lidar based on the initial global pose of the lidar According to t i right Perform linear interpolation to unify the positions of all points in each frame of point cloud to the start time of the frame point cloud, and remove the motion distortion in each frame of point cloud. Transform the points after removing motion distortion to the global coordinate system Rasterize the processed points and count the number of laser scanning points N in each grid. i , for N i Greater than a given threshold N τ Calculate the mean μ of all points in the three-dimensional grid i and covariance∑ i . Covariance∑ i Perform eigenvalue decomposition to obtain a feature matrix containing eigenvalues. Based on the feature matrix, construct a least squares constraint, where the motion increment is used as the initial value of the least squares constraint. Solve the least squares constraint and determine the first global pose.

[0064] According to the eigenvalues ​​of the feature matrix, the point cloud features in the grid are judged to be surface features, line features or Gaussian distribution features, and the eigenvalues ​​in the inverse matrix corresponding to the feature matrix are adjusted according to the judgment results; according to the inverse matrix corresponding to the adjusted feature matrix, the results of the eigenvalue decomposition of the covariance matrix are determined, and the least squares constraints are constructed based on the results.

[0065] For example, when the characteristic matrix is ​​a diagonal matrix, the characteristic matrix includes a first eigenvalue, a second eigenvalue, and a third eigenvalue, wherein the first eigenvalue is greater than the second eigenvalue, and the second eigenvalue is greater than the third eigenvalue. If the difference between the first eigenvalue and the second eigenvalue is greater than a first threshold, the point cloud feature in the grid is determined to be a line feature, and the fourth eigenvalue in the inverse matrix corresponding to the characteristic matrix is ​​set to 0, wherein the fourth eigenvalue and the first eigenvalue are reciprocals of each other; if the difference between the first eigenvalue and the second eigenvalue is less than or equal to the first threshold, and the difference between the second eigenvalue and the third eigenvalue is greater than the second threshold, the point cloud feature in the grid is determined to be a surface feature, and the fourth eigenvalue and the fifth eigenvalue in the inverse matrix corresponding to the characteristic matrix are set to 0, wherein the fifth eigenvalue and the second eigenvalue are reciprocals of each other; if the difference between the second eigenvalue and the third eigenvalue is less than or equal to the second threshold, the point cloud feature in the grid is determined to be a Gaussian distribution feature, and the fourth eigenvalue, the fifth eigenvalue, and the sixth eigenvalue in the inverse matrix corresponding to the characteristic matrix are not adjusted, wherein the sixth eigenvalue and the third eigenvalue are reciprocals of each other. The difference between the first eigenvalue and the second eigenvalue is greater than the first threshold, and the difference between the second eigenvalue and the third eigenvalue is greater than the second threshold, the first eigenvalue is much greater than the second eigenvalue, and the second eigenvalue is much greater than the third eigenvalue. For example, the first eigenvalue and the second eigenvalue are not of the same order of magnitude, and the second eigenvalue and the third eigenvalue are not of the same order of magnitude. For another example, the first eigenvalue is 10 times or more, 100 or more, etc. of the second eigenvalue, and the second eigenvalue is 10 times or more, 100 times or more, etc. of the third eigenvalue.

[0066] For example, i =U T ΛU, where three eigenvalues λ1, λ2, λ3 of the feature matrix Λ. If λ1 >> λ2, it is considered that the point cloud feature in the grid is a line feature, and the is set to 0. If λ2 >> λ3, it is considered that the point cloud feature in the grid is a surface feature, and the and are set to 0; if neither of the above two, it is considered that the point cloud feature in the grid is a Gaussian distribution feature, and the feature matrix Λ is not adjusted. Through the processing of the feature matrix Λ, the error of subsequent map construction can be reduced.

[0067] In some embodiments, the constructed least squares constraint is where R, P are motion information to be estimated, X i is a point in the new point cloud, N is the number of grids, w i is the weight of each grid, μ i is the grid mean, U T ΛU is the result of eigenvalue decomposition of the grid covariance.

[0068] By searching for the nearest neighbor grid of X i , the corresponding relationship can be found. The laser radar motion increment calculated according to the initial global pose of the laser radar is taken as the initial value of the least squares problem, and solving this least squares problem can obtain the pose of the current laser radar in the global coordinate system , that is, the first global pose of the laser radar after optimization.

[0069] In step 340, according to the first global pose, the point cloud is spliced to obtain a first submap, and a second submap adjacent to the first submap is registered according to the first global pose, and the pose transformation relationship between the first submap and the second submap is taken as a laser odometry factor.

[0070] In some embodiments, the de-distorted point clouds of the continuous N frames are spliced according to the first global pose to obtain a submap Submap i , and all point clouds in the submap Submap i are transformed into the coordinate system of the first frame of the submap Submap i ; the first global pose is used to register the adjacent two submaps Submap i and Submap i+1 in the first frame coordinate system to obtain the pose transformation relationship between the two submaps , which is taken as a laser odometry factor added to the factor graph model.

[0071] At step 350, loop closure detection is performed on the first submap as a loop closure detection unit to determine a loop closure detection candidate submap corresponding to the first submap.

[0072] In some embodiments, a plurality of historical submaps corresponding to the first submap are obtained; the plurality of historical submaps are sorted in chronological order to form a historical submap sequence; the first global pose is used to sequentially register the first submap and each historical submap in the historical submap sequence; the first submap and the registered historical submap are spliced, and all point pairs within a first preset distance in the first submap and the registered historical submap are found, and distance values between the all point pairs are calculated; if the distance values are less than a distance threshold, the registered historical submap is taken as a loop closure detection candidate submap corresponding to the first submap.

[0073] For example, after a submap is constructed, a screening region is constructed with a global position of a first frame of a laser radar coordinate system in the latest submap as a center and a second preset distance (for example, 15 m) as a radius, and historical submaps in the screening region are determined. After a plurality of historical submaps are searched, the historical submaps are sorted in chronological order to form a historical submap sequence. The first global pose is used to sequentially register the current submap and historical submaps in the historical submap sequence, wherein a registration initial value is the first global pose. The first global pose is used to sequentially register the current submap and historical submaps in the historical submap sequence, wherein a registration initial value is the first global pose. According to a registration result The two submaps are spliced, and all point pairs within a first preset distance (for example, 2 m) in the two submaps are found, and distance values (for example, average distance) between the all point pairs are calculated. When the distance values are not greater than a preset distance value (for example, 1), it is considered that a loop is successfully detected this time, and the historical submap is taken as a loop closure detection candidate submap, and loop closure detection is stopped. The current submap is taken as a loop closure detection candidate submap, and loop closure detection is stopped. Meanwhile, a pose transformation relationship between the current submap and the loop closure detection candidate submap

[0074] is taken as a loop closure detection factor and added to a factor graph. When the distance values are greater than a preset distance threshold, it is considered that loop closure detection fails, and the next loop closure detection is performed.

[0075] In the above steps, map layering can be effectively eliminated through loop closure detection.

[0076] In some embodiments, as Figure 4 ​As shown, after the new GPS prior factor or the loop closure detection factor is added to the factor graph model, an error is constructed with the laser odometry factor already existing in the factor graph model, and the global pose of the laser radar is corrected according to the error minimization principle to determine the second global pose of the laser radar coordinate system for further optimization. The laser odometry factor is generated and always exists in the factor graph model, that is, the laser odometry factor is a basic factor in the factor graph model.

[0077] In some embodiments, the GPS prior factor or the loop closure detection factor can be generated alone, or the GPS prior factor and the loop closure detection factor can exist simultaneously. That is, the factor graph model containing the laser odometry factor can only add the GPS prior factor or the loop closure detection factor, or can add the GPS prior factor and the loop closure detection factor simultaneously. When only the GPS prior factor is added to the factor graph model, the GPS prior factor constructs an error with the laser odometry factor already existing in the factor graph model; when only the loop closure detection factor is added to the factor graph model, the loop closure detection factor constructs an error with the laser odometry factor already existing in the factor graph model; when the GPS prior factor and the loop closure detection factor are added to the factor graph model simultaneously, the GPS prior factor and the loop closure detection factor construct an error with the laser odometry factor already existing in the factor graph model.

[0078] In some embodiments, since the loop closure detection factor and the GPS prior factor both correct the global pose of the laser radar, the corresponding factors are updated after the factor graph model is optimized. The GPS prior factor can effectively eliminate the cumulative error, so that the constructed three-dimensional map is aligned with the geographic coordinate system, and the mapping accuracy is improved.

[0079] In step 370, the plurality of first subgraphs are aligned to the global coordinate system, and the plurality of first subgraphs are spliced into a three-dimensional map according to the second global pose.

[0080] In some embodiments, indicates that the subgraph is first aligned to the global coordinate system, and then the subgraph is spliced.

[0081] In the above embodiment, the laser radar and the GNSS integrated inertial navigation are fused, and when applied to the unmanned mine truck, a high-precision three-dimensional map of the loading area in the mine area can be constructed as long as the laser radar can scan the road edge information. Compared with the prior art, on the one hand, the situation that the mine truck cannot collect road edge information when driving to a narrow area is avoided, and the integrity of the collected road edge information is ensured; on the other hand, the mine truck does not need to drive along the road edge to obtain all the road edge information, and under the premise of ensuring the construction accuracy, the pre-acquisition data work of the construction is simplified, the time consumption is short, and the overall work efficiency is improved; on the other hand, the road edge in the loading area in the mine area will change due to continuous construction work, and the mine truck can quickly obtain the changed road edge information according to the preset path in a short time through the technical solution provided by the present application. Compared with the two-dimensional map containing only the road edge, the present disclosure can construct a precise three-dimensional map of the loading area in the mine area, improve the perception and path planning capability of the mine truck, and accurately guide the unmanned mine truck to complete the loading operation.

[0082] In some embodiments of the present disclosure, the three-dimensional map in the global coordinate system is converted to the geographic coordinate system.

[0083] In the above embodiment, the three-dimensional map is constructed in the global coordinate system, which creates favorable conditions for converting the three-dimensional map to the geographic coordinate system, and helps to further improve the mapping accuracy.

[0084] Figure 5 A structural schematic diagram of some embodiments of the three-dimensional mapping device of the unmanned mine area of the present disclosure. The three-dimensional mapping device includes a splicing module 510, a laser odometry factor acquisition module 520, and a mapping module 530.

[0085] The splicing module 510 is configured to optimize the initial global pose of the laser radar to obtain a first global pose according to the point cloud features obtained by the laser radar located on the mine truck, and splice the point cloud to obtain a first sub-map according to the first global pose.

[0086] In some embodiments, the splicing module 510 calculates the relative rotation time of each point for each frame of point cloud obtained by the laser radar; calculates the motion increment of the laser radar based on the initial global pose; linearly interpolates the motion increment using the relative rotation time to unify the positions of all points in each frame of point cloud to the time when the frame of point cloud starts; converts the processed points to the global coordinate system; performs rasterization processing on the processed points, counts the number of points in each grid, and if the number of points in the grid is greater than a number threshold, calculates the mean and covariance of all points in the grid, performs eigenvalue decomposition on the covariance matrix to obtain a feature matrix containing eigenvalues; based on the feature matrix, a least squares constraint is constructed, wherein the motion increment is taken as the initial value of the least squares constraint, the least squares constraint is solved, and the first global pose is determined.

[0087] In some embodiments, the stitching module 510 determines whether the point cloud features in the grid are surface features, line features, or Gaussian distribution features based on the eigenvalues ​​of the feature matrix, and adjusts the eigenvalues ​​in the inverse matrix corresponding to the feature matrix based on the judgment result; determines the result of the eigenvalue decomposition of the covariance matrix based on the inverse matrix corresponding to the adjusted feature matrix, and constructs the least squares constraint based on the result.

[0088] For example, ∑ i =U T ΛU, where For the three eigenvalues ​​λ1, λ2, and λ3 of the feature matrix Λ. If λ1>>λ2, the point cloud features in the grid are considered to be line features, and the Set to 0. If λ2>>λ3, the point cloud features in the grid are considered to be surface features, and the feature matrix Λ and Both are set to 0; if neither of the above two is true, the point cloud features in the grid are considered to be Gaussian distribution features, and the feature matrix Λ is not adjusted.

[0089] In some embodiments, the least squares constraint is constructed as Among them, R, P are the motion information to be estimated, X i is a point in the new point cloud, N is the number of grids, w i The weight of each grid, μ i is the grid mean, U T ΛU is the result of grid covariance eigenvalue decomposition.

[0090] In some embodiments, the stitching module 510 stitches the point clouds after dedistortion of N consecutive frames according to the first global pose to obtain a first sub-image.

[0091] The laser odometry factor acquisition module 520 is configured to register the second sub-image adjacent to the first sub-image according to the first global pose, and use the pose transformation relationship between the first sub-image and the second sub-image as the laser odometry factor.

[0092] In some embodiments, the laser odometry factor acquisition module 520 uses the first global pose to align the first sub-image and the second sub-image in the first frame coordinate system of the first sub-image, obtains the pose transformation relationship between the first sub-image and the second sub-image, and uses the pose transformation relationship as the laser odometry factor.

[0093] The mapping module 530 is configured to optimize the first global pose according to errors of a global positioning system (GPS) prior factor and a loop closure detection factor and a laser odometry factor, to obtain a second global pose, wherein the GPS prior factor is displacement information in the initial global pose, and the loop closure detection factor is a pose transformation relationship between the first submap and a loop closure detection candidate submap corresponding to the first submap, align the plurality of first submaps to a global coordinate system, and splice the plurality of first submaps into the three-dimensional map according to the second global pose.

[0094] In some embodiments, the mapping module 530 is further configured to convert the three-dimensional map in the global coordinate system to a geographic coordinate system.

[0095] In the above embodiments, the global pose of the lidar is optimized according to errors of the GPS prior factor and the loop closure detection factor and the laser odometry factor, and then the plurality of submaps are spliced into the three-dimensional map based on the global pose, that is, a more accurate and comprehensive three-dimensional map can be constructed through multi-factor optimization and multi-level optimization, and the problems of map degradation and error accumulation in the three-dimensional map modeling process are reduced.

[0096] In some embodiments of the present disclosure, as shown in Figure 6 The three-dimensional mapping device further includes an external parameter acquisition module 610 and a GPS prior factor acquisition module 620.

[0097] The external parameter acquisition module 610 is configured to acquire an external parameter between the lidar and a center of a rear axle of a vehicle body of the mine truck, and acquire global positioning information of the center of the rear axle of the vehicle body by using a GNSS integrated inertial navigation system located on the mine truck.

[0098] The GPS prior factor acquisition module 620 is configured to acquire an initial global pose of the lidar according to the external parameter and the global positioning information, and take displacement information in the initial global pose as a GPS prior factor.

[0099] The three-dimensional mapping device further includes a loop closure detection factor acquisition module 630 configured to perform loop closure detection by taking the first submap as a loop closure detection unit, and determine a loop closure detection candidate submap corresponding to the first submap.

[0100] In some embodiments, the loop closure detection factor acquisition module 630 acquires a plurality of historical submaps corresponding to the first submap, sorts the plurality of historical submaps in chronological order to form a historical submap sequence, sequentially performs registration on the first submap and each historical submap in the historical submap sequence by using the first global pose, splices the first submap and the registered historical submap, finds all point pairs within a first preset distance in the first submap and the registered historical submap, and calculates distance values between the all point pairs, and if the distance values are less than a distance threshold, the registered historical submap is taken as the loop closure detection candidate submap corresponding to the first submap.

[0101] In some embodiments, a screening region is constructed with the first frame lidar coordinate system global position in the first submap as the center and a second preset distance as the radius, and a plurality of historical submaps in the screening region are determined.

[0102] In the above embodiment, in view of the open feature of the loading area in the mining area, a plurality of features of the point cloud are adaptively extracted, and a least square constraint is constructed based on the plurality of features of the point cloud to optimize the global pose of the lidar, thereby solving the degradation problem of the traditional lidar mapping scheme in the open scene of the loading area and ensuring the mapping accuracy.

[0103] In addition, the present disclosure adopts a hierarchical optimization strategy to continuously optimize the map at multiple levels. For example, first, the point clouds between frames are spliced, and on this basis, a submap is constructed and the submap is registered to further align the point clouds and use the registration result of the submap as a laser odometry factor, and a GPS factor is added, wherein the GPS factor can effectively eliminate the cumulative error, so that the constructed three-dimensional map is aligned with the geographic coordinate system, and the mapping accuracy is improved. Then, the submap is taken as a unit to perform loop closure detection to obtain a loop closure detection factor, and the GPS factor, the laser odometry factor and the loop closure detection factor are added to a factor graph model to realize global pose optimization of the lidar, and the mapping accuracy is further improved.

[0104] Figure 7 The structure diagram of another embodiment of the three-dimensional mapping device of the unmanned mining area of the present disclosure. The three-dimensional mapping device 700 includes a memory 710 and a processor 720. Among them: the memory 710 can be a disk, a flash memory or any other non-volatile storage medium. The memory 710 is used to store the instructions in the above embodiments. The processor 720 is coupled to the memory 710 and can be implemented as one or more integrated circuits, such as a microprocessor or a microcontroller. The processor 720 is used to execute the instructions stored in the memory.

[0105] In some embodiments, the processor 720 is coupled to the memory 710 through the BUS bus 730. The three-dimensional mapping device 700 can also be connected to an external storage system 750 through a storage interface 740 to call external data, and can also be connected to a network or another computer system (not shown) through a network interface 760. Details are not described here.

[0106] In this embodiment, the data instructions are stored in the memory, and the above instructions are processed by the processor, which can improve the mapping accuracy and solve the problems of map degradation and error accumulation in the three-dimensional map modeling process.

[0107] In some embodiments, a non-transitory computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the steps of the methods of the corresponding embodiments described above. Those skilled in the art should appreciate that embodiments of the present disclosure can be provided as methods, apparatuses, or computer program products. Accordingly, the present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure can take the form of a computer program product on one or more computer readable storage media (including, but not limited to, disk memory, CD-ROMs, optical storage devices, etc.) embodying computer readable program code thereon for execution by the computer.

[0108] The present disclosure is described in reference to flowcharts and / or block diagrams of methods, apparatuses (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, a special purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions, which are executed via the processor of the computer or other programmable data processing apparatus, generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in the flowcharts and / or block diagrams.

[0109] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufacture product including an instruction apparatus that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in the flowcharts and / or block diagrams.

[0110] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in the flowcharts and / or block diagrams.

[0111] So far, the present disclosure has been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Those skilled in the art can fully understand how to implement the technical solutions disclosed herein according to the above description.

[0112] While certain embodiments of the disclosure have been described herein in detail as presently preferred, many modifications and variations thereof will be apparent to those skilled in the art, without departing from the scope and spirit of the disclosure. It is to be understood that those skilled in the art will be able to devise many embodiments of the disclosure which, while not explicitly described or shown herein, embody the principles of the disclosure and are included within its spirit and scope. Accordingly, all such suitable modifications and equivalents should be considered as within the scope of the disclosure. The scope of the disclosure is to be indicated by the appended claims, rather than the foregoing description, and all changes that come within the meaning and range of equivalents are intended to be embraced therein.

Claims

1. A 3D mapping method for an unmanned mining area, comprising: Optimizing an initial global pose of the laser radar based on point cloud features acquired by a laser radar on the mining car to obtain a first global pose, and stitching the point cloud based on the first global pose to obtain a first sub-image; Acquire multiple historical subgraphs corresponding to the first subgraph, sort the multiple historical subgraphs in chronological order to form a historical subgraph sequence, sequentially register the first subgraph with each historical subgraph in the historical subgraph sequence using the first global pose, splice the first subgraph with the registered historical subgraphs, find all point pairs within a first preset distance between the first subgraph and the registered historical subgraphs, calculate the distance between all point pairs, and if the distance value is less than a distance threshold, use the registered historical subgraph as a candidate subgraph for loop closure detection corresponding to the first subgraph; According to the first global pose, register a second sub-image adjacent to the first sub-image, and use the pose transformation relationship between the first sub-image and the second sub-image as a laser odometry factor; Optimizing the first global pose based on an error between at least one of a Global Positioning System (GPS) prior factor and a closed-loop detection factor and the laser odometry factor to obtain a second global pose, wherein the GPS prior factor is displacement information in the initial global pose, and the closed-loop detection factor is a pose transformation relationship between the first subgraph and a closed-loop detection candidate subgraph corresponding to the first subgraph; as well as The plurality of first sub-images are aligned to a global coordinate system, and the plurality of first sub-images are spliced ​​into a three-dimensional map according to the second global pose.

2. The 3D mapping method according to claim 1, further comprising: Obtaining external parameters between the laser radar and the center of the rear axle of the mining car, and obtaining global positioning information of the center of the rear axle of the mining car using a global navigation satellite system (GNSS) combined inertial navigation system located on the mining car; as well as Obtain an initial global pose of the laser radar based on the external parameters and the global positioning information.

3. The three-dimensional mapping method according to claim 1, wherein: Acquiring multiple historical subgraphs corresponding to the first subgraph includes: A screening area is constructed with the global position of the first frame laser radar coordinate system in the first sub-image as the center and the second preset distance as the radius, and multiple historical sub-images within the screening area are determined.

4. The 3D mapping method for an unmanned mining area according to any one of claims 1 to 3, wherein: Obtaining the first global pose includes: For each frame of point cloud acquired by the laser radar, calculating the relative rotation time of each point; Calculating a motion increment of the laser radar based on the initial global pose; Performing linear interpolation on the motion increment using the relative rotation time to unify the positions of all points in each frame of point cloud to the start time of the frame of point cloud; Convert the processed points to the global coordinate system; Performing gridding on the processed points, counting the number of points in each grid, and if the number of points in the grid is greater than a threshold, calculating the mean and covariance of all points in the grid, performing eigenvalue decomposition on the covariance matrix, and obtaining a feature matrix containing eigenvalues; and Based on the characteristic matrix, a least squares constraint is constructed, wherein the motion increment is used as an initial value of the least squares constraint, the least squares constraint is solved, and the first global pose is determined.

5. The three-dimensional mapping method according to claim 4, wherein: Based on the feature matrix, constructing the least squares constraint includes: Determining whether the point cloud features within the grid are surface features, line features, or Gaussian distribution features based on the eigenvalues ​​of the feature matrix, and adjusting the eigenvalues ​​in the inverse matrix corresponding to the feature matrix based on the determination result; and The result of eigenvalue decomposition of the covariance matrix is ​​determined according to the inverse matrix corresponding to the adjusted characteristic matrix, and the least squares constraint is constructed according to the result.

6. The three-dimensional mapping method according to claim 5, wherein: Judging whether the point cloud feature in the grid is a surface feature, a line feature, or a Gaussian distribution feature according to the eigenvalues ​​of the feature matrix, and adjusting the eigenvalues ​​in the inverse matrix corresponding to the feature matrix according to the judgment result includes: In the case where the characteristic matrix is ​​a diagonal matrix, the characteristic matrix includes a first eigenvalue, a second eigenvalue, and a third eigenvalue, wherein the first eigenvalue is greater than the second eigenvalue, and the second eigenvalue is greater than the third eigenvalue. If the difference between the first eigenvalue and the second eigenvalue is greater than a first threshold, determining that the point cloud feature in the grid is a line feature, and setting the fourth eigenvalue in the inverse matrix corresponding to the feature matrix to 0, wherein the fourth eigenvalue and the first eigenvalue are reciprocals of each other; If the difference between the first eigenvalue and the second eigenvalue is less than or equal to a first threshold, and the difference between the second eigenvalue and the third eigenvalue is greater than a second threshold, determining that the point cloud feature in the grid is a surface feature, and setting the fourth eigenvalue and the fifth eigenvalue in the inverse matrix corresponding to the feature matrix to 0, wherein the fifth eigenvalue and the second eigenvalue are reciprocals of each other; and If the difference between the second eigenvalue and the third eigenvalue is less than or equal to a second threshold, it is determined that the point cloud feature in the grid is a Gaussian distribution feature, and the fourth eigenvalue, the fifth eigenvalue and the sixth eigenvalue in the inverse matrix corresponding to the feature matrix are not adjusted, wherein the sixth eigenvalue and the third eigenvalue are reciprocals of each other.

7. The three-dimensional mapping method according to any one of claims 1 to 3, wherein: According to the first global pose, stitching the point cloud to obtain a first sub-image includes: The point clouds after dedistortion of N consecutive frames are stitched according to the first global pose to obtain the first sub-image.

8. The three-dimensional mapping method according to any one of claims 1 to 3, wherein: According to the first global pose, registering a second sub-image adjacent to the first sub-image, and using the pose transformation relationship between the first sub-image and the second sub-image as a laser odometry factor includes: Using the first global pose, the first sub-image and the second sub-image are aligned in the first frame coordinate system of the first sub-image to obtain a pose transformation relationship between the first sub-image and the second sub-image, and the pose transformation relationship is used as a laser odometry factor.

9. The 3D mapping method according to any one of claims 1 to 3, further comprising: The three-dimensional map in the global coordinate system is converted to the geographic coordinate system.

10. A 3D mapping device for an unmanned mining area, comprising: a stitching module configured to optimize an initial global pose of the laser radar based on point cloud features acquired by the laser radar on the mining car to obtain a first global pose, and stitch the point cloud based on the first global pose to obtain a first sub-image; a closed-loop detection factor acquisition module configured to acquire multiple historical subgraphs corresponding to the first subgraph, sort the multiple historical subgraphs in chronological order to form a historical subgraph sequence, sequentially register the first subgraph with each historical subgraph in the historical subgraph sequence using the first global pose, concatenate the first subgraph with the registered historical subgraphs, find all point pairs within a first preset distance between the first subgraph and the registered historical subgraphs, calculate distance values ​​between all point pairs, and if the distance value is less than a distance threshold, use the registered historical subgraph as a candidate subgraph for closed-loop detection corresponding to the first subgraph; a laser odometry factor acquisition module configured to register a second sub-image adjacent to the first sub-image according to the first global pose, and use a pose transformation relationship between the first sub-image and the second sub-image as a laser odometry factor; as well as A mapping module is configured to optimize the first global pose based on an error between at least one of a global positioning system (GPS) prior factor and a closed-loop detection factor and the laser odometry factor to obtain a second global pose, wherein the GPS prior factor is the displacement information in the initial global pose, and the closed-loop detection factor is the pose transformation relationship between the first subgraph and a closed-loop detection candidate subgraph corresponding to the first subgraph, align the plurality of first subgraphs to a global coordinate system, and splice the plurality of first subgraphs into a three-dimensional map according to the second global pose.

11. The 3D mapping device according to claim 10, further comprising: an external parameter acquisition module, configured to acquire external parameters between the laser radar and the center of the rear axle of the mining car, and to acquire global positioning information of the center of the rear axle of the mining car using a global navigation satellite system (GNSS) combined inertial navigation system located on the mining car; as well as The GPS prior factor acquisition module is configured to obtain the initial global pose of the laser radar based on the external parameters and the global positioning information, and use the displacement information in the initial global pose as the GPS prior factor.

12. A 3D mapping device for an unmanned mining area, comprising: Memory; as well as A processor coupled to the memory, wherein the processor is configured to execute the three-dimensional mapping method according to any one of claims 1 to 9 based on instructions stored in the memory.

13. A non-transitory computer-readable storage medium having computer program instructions stored thereon, wherein when the instructions are executed by a processor, the three-dimensional mapping method according to any one of claims 1 to 9 is implemented.

Images