Three-dimensional point cloud map construction method and system

Abstract

The embodiments of the present application provide a three-dimensional point cloud map construction method and system for robot three-dimensional point cloud map construction and mapping tasks of unknown scenes. The existing three-dimensional space mapping scheme for large artificial scenes randomly selects target points when planning walking routes, which has an adverse effect on the mapping effect. The present application uses multi-line laser radar and SLAM algorithm to obtain the three-dimensional spatial structure information of the scene; then, the best target point is selected based on the expected maximization of information gain; then, an exploration path from the current position to the best target point is generated based on the three-dimensional map information and the three-dimensional morphology of the robot; finally, tracking navigation movement is performed based on the exploration path. As the map information continues to change, the robot selects a new target point based on the updated map, and iterates repeatedly until the two-dimensional grid map of the target map forms a closed explored space and outputs a global three-dimensional point cloud map.

Description

Technical Field

[0001] The present application relates to a three-dimensional point cloud map construction method and system, belonging to the intersection of robotics and measurement technology. Background Art

[0002] When existing robots are deployed in actual work scenarios, especially in large-scale scenarios such as large shopping malls, underground garages, civil air defense projects, underground mine exploration, and rescue in dangerous areas, they usually need to use multi-line lidar to obtain three-dimensional (3D) spatial information of the scene, so as to realize various downstream application tasks, such as security inspection, cleaning, disinfection, distribution, building surveying and mapping, settlement estimation, automatic positioning, etc.

[0003] At present, robots used to build three-dimensional maps are divided into two categories according to the control method. One is controlled by remote control, which controls the robot's walking direction and speed in real time through the remote control. The other is autonomous control robots, which can automatically collect scene information and automatically plan walking paths based on the scene information. The entire three-dimensional map construction process does not require human control intervention. Autonomous control robots are popular in the market because they can save a lot of labor costs.

[0004] However, when an autonomous control robot plans a walking path, it first needs to determine a target point. Existing robots mostly select target points by randomly selecting from multiple points that meet certain conditions. The target point selected in this way is not necessarily the best among all points. Therefore, the method of randomly selecting target points has an adverse effect on the effect of the robot in constructing a three-dimensional map to a certain extent. Summary of the invention

[0005] In order to solve one of the above technical defects, the embodiment of the present application provides a method and system for constructing a three-dimensional point cloud map, which uses a three-dimensional multi-line laser radar and a mobile robot or an unmanned vehicle to autonomously explore and map unknown scenes; at the same time, it is not affected by ambient light, and finally obtains a three-dimensional space point cloud map within the robot's accessible area (Achievable Region). The whole method includes four steps: 1) local three-dimensional point cloud map construction and mileage estimation based on SLAM (Simultaneous Localization And Mapping); 2) target point search based on information gain; 3) exploration path generation based on collision avoidance; 4) robot tracking control.

[0006] According to a first aspect of an embodiment of the present application, a method for constructing a three-dimensional point cloud map is provided, which specifically includes:

[0007] Obtain point cloud and construct a local three-dimensional point cloud map based on the point cloud;

[0008] Determine the exploration gain at the target point based on the local 3D point cloud map, and determine the best target point based on the exploration gain;

[0009] Plan the exploration path and drive the robot to move along the exploration path;

[0010] The local three-dimensional point cloud map is updated according to the point cloud obtained by the robot to obtain the global three-dimensional point cloud map.

[0011] According to a second aspect of an embodiment of the present application, a three-dimensional point cloud map construction system is provided, specifically comprising:

[0012] A map construction module is used to obtain point clouds and construct a local three-dimensional point cloud map based on the point clouds;

[0013] A target determination module is used to determine the exploration gain at the target point according to the local three-dimensional point cloud map, and determine the best target point according to the exploration gain;

[0014] The target exploration module is used to plan the exploration path and drive the robot to move along the exploration path;

[0015] The map updating module is used to update the local three-dimensional point cloud map according to the point cloud acquired by the robot to obtain a global three-dimensional point cloud map.

[0016] The three-dimensional point cloud map construction method and system provided in the embodiments of the present application have the following advantages:

[0017] 1. Based on the two-dimensional grid map information of the two-dimensional reachable area and the target point screening strategy based on multi-target information gain, the exploration point selection is performed to give the robot the ability to understand the semantic information of the unknown environment, and constrain the robot to complete the local search first, prevent frequent changes in the exploration direction, and reduce exploration jitter. And through the information gain inspiration strategy of the overall map, the exploration efficiency is improved.

[0018] 2. Long-distance scene measurement (>100m) is achieved through multi-line laser radar, and accurate spatial three-dimensional information (building structures and obstacles) and precise odometer information are obtained based on three-dimensional SLAM to replace the robot chassis odometer, avoiding mileage estimation errors caused by uneven road surfaces and water accumulation, making the robot positioning more accurate and the operation safer.

[0019] 3. Trajectory generation based on local collision avoidance uses the “storage for efficiency” approach to calculate multiple potentially feasible trajectories based on local map information at one time, and organizes them in a hierarchical manner, thereby reducing the overall computational load of the robot. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0021] Figure 1 A flowchart of a method for constructing a three-dimensional point cloud map provided in an embodiment of the present application;

[0022] Figure 2 Schematic diagram of a three-dimensional test simulation scene of the three-dimensional point cloud map construction method provided in an embodiment of the present application, wherein (a) is a small indoor scene, and (b) is a large basement scene;

[0023] Figure 3 Schematic diagram of a map provided in an embodiment of the present application, wherein (a) is a local three-dimensional point cloud map, and (b) is a two-dimensional grid map;

[0024] Figure 4 Schematic diagram of boundary points provided in an embodiment of the present application, wherein (a) is a schematic diagram of boundary points on a two-dimensional grid map, and (b) is a schematic diagram of boundary points on a three-dimensional point cloud map;

[0025] Figure 5 A schematic diagram of the main path and candidate paths provided in the embodiment of the present application;

[0026] Figure 6 A schematic diagram showing the growth of the area of ​​the area explored by the robot during the exploration process provided by the embodiment of the present application;

[0027] Figure 7 A schematic diagram of a three-dimensional point cloud map obtained after the large-scale basement scene exploration provided in an embodiment of the present application;

[0028] Figure 8 A schematic diagram of a three-dimensional point cloud map obtained after the small indoor scene exploration provided in an embodiment of the present application;

[0029] Fig. 9 Schematic diagram of exploration time for performing ten explorations at different locations in a small indoor scene and a large basement scene provided in an embodiment of the present application, wherein (a) is a schematic diagram of exploration time in a small indoor scene, and (b) is a schematic diagram of exploration time in a large basement scene;

[0030] Fig.10 Schematic diagrams of the robot provided in the embodiments of the present application in different initial positions, wherein (a) is a schematic diagram of the initial position in a small indoor scene, and (b) is a schematic diagram of the initial position in a large basement scene. DETAILED DESCRIPTION

[0031] In order to make the technical solutions and advantages in the embodiments of the present application more clearly understood, the exemplary embodiments of the present application are further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than an exhaustive list of all the embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0032] In the process of implementing this application, the applicant discovered that the target points selected by the existing autonomous control robots when planning the walking path are random, that is, a point is randomly selected as the target point from multiple candidate points that meet the conditions, and then the walking path between the robot's current detection point and the target point is planned. For example, the technical solution disclosed in patent application CN112964263A discloses an automatic mapping method, a mobile robot and a readable storage medium. The patent application randomly selects a point on a specific fan-shaped edge as the target point, and then plans the target path between the current detection point and the target point. Although all points in the candidate points can be used as target points, each candidate point has a different impact on the subsequent mapping effect, so the method of randomly selecting target points cannot achieve the best mapping effect.

[0033] In response to the above problems, the embodiments of the present application provide a method, system, electronic device and computer storage medium for constructing a three-dimensional point cloud map, the method comprising: calibrating the external parameters of a multi-line laser radar, determining the conversion relationship between the robot body coordinate system and the multi-line laser radar coordinate system; obtaining continuous frame point clouds, and simultaneously performing radar mileage estimation and construction of a local three-dimensional point cloud map; determining the best target point for the robot, and driving the robot to explore the best target point; generating an exploration path, and driving the robot to explore along the exploration path; and updating the global three-dimensional point cloud map. Using the method of the present application, the robot can automatically find the target point with the greatest exploration benefit, and plan the exploration path, and can obtain the best mapping effect.

[0034] In the embodiments of the present application, a method for constructing a three-dimensional point cloud map is provided, such as Figure 1 As shown, the method includes:

[0035] S100, calibrating the external parameters of the multi-line laser radar to determine the conversion relationship between the robot body coordinate system and the multi-line laser radar coordinate system.

[0036] Exemplarily, a calibration object can be set in front of the robot chassis, and then a multi-line laser radar can be used to measure the calibration object. The conversion relationship between the robot body coordinate system and the laser radar coordinate system can be calculated through the measurement value of the multi-line laser radar. Then, the installation height of the multi-line laser radar is measured to translate the Z-axis origin of the point cloud acquired by the multi-line laser radar to the ground for height-based initial ground filtering. Specifically, after the multi-line laser radar collects the ground point cloud, the height threshold is used to crop the ground point cloud, and the RANSAC (Random Sample Consensus) algorithm is used to perform plane fitting on the cropped ground point cloud to obtain the precise installation height of the multi-line laser radar.

[0037] S120, obtaining continuous frame point clouds, and simultaneously performing radar mileage estimation and construction of a local three-dimensional point cloud map.

[0038] For example, the multi-line laser radar collects point clouds in the scene at set time intervals, and the point clouds at two adjacent time points t and t+1 are called continuous frame point clouds. After obtaining the continuous frame point clouds, the continuous frame point clouds {P1, P2, …, P n} Perform local 3D point cloud mapping and mileage estimation, for the point cloud P of two adjacent frames t and P t+1 , respectively calculate the edge feature P based on the following smoothness formula (1) t E and P t+1 E , surface feature P t p and P t+1 p .

[0039]

[0040] Among them, c represents the smoothness of the point cloud, S is the neighborhood range of the current point cloud, and r j represents the Euclidean distance from the point cloud with index j to the sensor origin, r i represents the Euclidean distance from the point cloud with index i to the sensor origin, |S| represents the number of point clouds in the neighborhood, and ||.|| represents the Euclidean norm.

[0041] Specifically, when determining edge features and face features, the RANSAC algorithm is first used to iteratively fit the point cloud to a plane, and the N largest fitting planes obtained by fitting are selected. If the number of points in the fitting plane is greater than 50% of the total number of points in the point cloud, the plane fitting is stopped. The points in the fitting plane are taken as plane feature point clouds, and the remaining points are taken as edge feature point clouds. The smoothness of the point cloud is calculated using formula (1), and then the smoothness is sorted. The plane feature point clouds with the highest sorting (highest smoothness) are taken as face features, and the edge feature point clouds with the lowest sorting (lowest smoothness) are taken as edge features.

[0042] Multi-line laser radar includes solid-state laser radar and surround laser radar. The surround laser radar rotates 360 degrees during operation to obtain point clouds in any direction. However, after the surround radar rotates, the point clouds obtained for the same object will have errors. Therefore, after obtaining the surface features, it is necessary to t p , P t+1 p The point cloud obtained by the surround-looking laser radar is registered, and then a local three-dimensional point cloud map is generated based on the registered point cloud, and the three-dimensional pose information T = [R, t] of each frame of the point cloud is estimated, where R = [roll, pitch, yaw] are the roll angle, pitch angle and heading angle of the multi-line laser radar, and t = [x, y, z] are the coordinates of the multi-line laser radar in three directions in the spatial coordinate system.

[0043] Specifically, when registering the point cloud obtained by the surround-view laser radar, it is necessary to use Lie group algebra SE (3) to calculate the point cloud P t+1 To P t T t+1 Specifically, we use the T matrix of the previous frame point cloud. t The matrix is ​​used as the initial solution to bring in the edge features {P t E , P t+1 E} set, and control the z direction and rotation parameters [roll, pitch] of the multi-line laser radar in the spatial coordinate system to remain unchanged, solve the x, y and heading angle yaw of the multi-line laser radar in the spatial coordinate system, and obtain T t+1 Matrix, using this T t+1 The matrix can be used for registration. Based on the updated T t+1 The matrix transforms the point cloud P t+1 Project to the local 3D point cloud map and perform voxel downsampling on the local 3D point cloud map (voxel unit = 0.1m 3 ), keep the overall point cloud density at a constant value. At the same time, based on T t+1Matrix inference robot's current position Tc = T1*T2,…,T t+1 .

[0044] In the embodiment of the present application, for the convenience of testing, the present application constructs two three-dimensional simulation scenes of different scales for effect verification. The scenes are as follows: Figure 2 As shown, Figure 2 (a) is a small indoor scene. Figure 2 (b) is a large basement scene. The continuous frame point cloud sequence obtained by the robot is {P1,P2,…,P n}, the smoothness of each frame point cloud is calculated by formula (1), and then the edge features and surface features are extracted. Then the relative pose transformation relationship [R, t] is solved through the point clouds of two adjacent frames. Then the point cloud of the current frame is projected to the global coordinate system through the relative pose transformation relationship, inserted into the acquired local 3D point cloud map, and voxelized and downsampled. Repeating this process can cyclically update the local 3D point cloud map. The local 3D point cloud map is as follows: Figure 3 As shown in (a) in .

[0045] S140, determining the best target point for the robot, and driving the robot to explore the best target point.

[0046] Exemplarily, first, the local three-dimensional point cloud map obtained by S120 is highly filtered to retain the point clouds within the robot's passable height (for example, the robot's chassis height) and the ground range. Then, the ground point cloud is obtained by fitting the RANSAC algorithm in combination with the ground normal vector constraint, and the point clouds outside the ground point cloud are non-ground point clouds. Then, all point clouds are projected onto the plane along the gravity direction. The projection height of the ground point cloud is 0, and the ground point cloud is also the area detected by the robot. The projection height of the non-ground point cloud is 1, and the non-ground point cloud is also the obstacle area. The projection height of other areas is -1, and other areas are also unknown areas that have not been detected by the robot. In this way, a two-dimensional grid map can be generated, such as Figure 3 As shown in (b).

[0047] After obtaining the above two-dimensional grid map, the best target point for the robot to explore can be determined. Specifically, based on the following gain formula (2), the target point with the largest exploration gain in the current two-dimensional grid map is estimated as the next exploration target, i.e., the best target point.

[0048] R(x p )=Gain(x p )×gainscale+Dis(x p )×disscale+Dir(x p )(2)

[0049] Among them, R represents the exploration benefit at the target point p, Gain is the information gain that the target point can obtain, which is represented by the area of ​​the unknown area that the sensor of the target point can cover, gainscale is its weight, which is used to reward those who can obtain greater information gain, Dis is the Euclidean distance between the robot and the target point, disscale is its weight, which is used to punish the target point farther away from the robot and encourage the exploration of nearby target points, Dir is the heading deviation between the target point direction and the robot body direction, which is used to encourage the robot to perform fewer steering operations, thereby reducing the backtracking phenomenon during the exploration process. The exploration benefit of the candidate target point is obtained by weighted calculation of these three indicators.

[0050] In formula (2), information gain Where Gain(p) is the information gain generated by the current two-dimensional grid map when the robot moves to the target point p, α is the current motion action (forward / left turn / right turn), Where θ is the target rotation angle.

[0051] Finally, the potential target points are clustered through hierarchical clustering to form multiple clusters, and the comprehensive exploration benefit of the current cluster is calculated based on the cluster. Where r is the radius of the minimum circumscribed circle covering the current cluster, R i represents the exploration benefit of the i-th target point in the cluster. The cluster with the largest comprehensive exploration benefit among all clusters is selected as the target cluster, and the target point closest to the cluster center is used as the optimal target point S to drive the robot to explore.

[0052] In the embodiment of the present application, a dense boundary point search is performed based on a two-dimensional grid map, and the point closest to the robot body in each cluster of continuous boundary points is extracted as a candidate target point. Figure 4 As shown in (a), the dots in the figure are candidate target points. The exploration benefit of each candidate target point is calculated by combining the information gain and the vehicle body distance through formula (2), and the target point with the largest exploration benefit is selected as the next exploration target point, that is, the optimal target point. Figure 4 (b) is a schematic diagram of the position of the corresponding optimal target point in the three-dimensional point cloud map.

[0053] S160: Generate an exploration path, and drive the robot to explore along the exploration path.

[0054] Exemplarily, based on the current position S of the robot and the position D of the best target point, the A* algorithm is used to search for a global main path that the robot can reach the best target point. Then, based on the global main path, a local candidate path set is generated in combination with the cubic spline curve of the robot chassis dynamics. The generated local candidate path set requires the robot to avoid collision with obstacles, and in the process of generating the local candidate path set, the generation range of the path can be controlled by limiting the total length of the local candidate path. At the same time, hierarchical path generation is performed in a recursive form, where the global main path is L0, and when generating the local candidate path L1, L0 is used as a constraint to generate several local candidate paths starting from L0 and ending at the L0 path; when generating the local candidate L2, the path in L1 is used as a constraint to generate several local candidate paths starting from L1 and ending at L1 or L0; and so on, the paths do not overlap during the generation process. The scale of the overall generation result and the amount of algorithm calculation are controlled by controlling the number of local candidate paths generated at each level and the number of levels.

[0055] In the embodiments of the present application, Figure 5 As shown. According to the obstacles perceived by the multi-line laser radar, the paths within the obstacle influence range are blocked, the probability of reaching the destination is calculated for the remaining paths, and the trajectory with the highest probability is selected for execution. Trajectory tracking navigation is performed based on the currently selected trajectory. When there are no obstacles during the driving process, the optimal trajectory towards the best target point is always selected for driving; when the selected trajectory is blocked by a new obstacle, any switchable path is selected for switching, and finally the target is reached.

[0056] S180: Update the local three-dimensional point cloud map to obtain a global three-dimensional point cloud map.

[0057] For example, the robot moves to the target point according to the selected trajectory, and the multi-line laser radar continuously updates the global 3D point cloud map and boundary point information of the scene during the process. After arriving at the target point, the next best target point is selected and navigated according to S140, and this process is repeated until there are no boundary points in the global 3D point cloud map, and the construction of the global 3D point cloud map is completed. Figure 6 During one exploration, as the robot continues to move and scan, the explored area gradually increases and the exploration is completed in about 550 seconds. The global 3D point cloud map constructed by the robot in the real scene is as follows: Figure 7 and Figure 8 shown.

[0058] To further test the effectiveness of the method in this application, Figure 2 The two simulation scenarios shown in the figure randomly select 5 different initial points for statistical testing. The initial selected positions are as follows: Fig.10The robot randomly starts from each point to explore and build a map. Each initial point is repeated 10 times. The total time to complete the robot's exploration is statistically analyzed. The resulting figure is shown in Fig. 9 As shown. Among them, for small indoor scenes, the robot basically completes the exploration in about 200-450s, and the 5 randomly selected points have little effect on the overall exploration time. The exploration statistics for large underground scenes show that the time required for robot exploration is between 1200-2700s, with large fluctuations. The median exploration time of the 5 randomly selected points is basically around 1800s, indicating that in the case of irregular large scenes, the algorithm exploration efficiency is basically stable, and it can complete the exploration autonomously without human operation, and the overall exploration time is also within the range allowed by the robot's battery life.

[0059] The embodiment of the present application also provides a three-dimensional point cloud map construction system, the system comprising:

[0060] A map construction module is used to obtain point clouds and construct a local three-dimensional point cloud map based on the point clouds;

[0061] A target determination module is used to determine the exploration gain at the target point according to the local three-dimensional point cloud map, and determine the best target point according to the exploration gain;

[0062] The target exploration module is used to plan the exploration path and drive the robot to move along the exploration path;

[0063] The map update module is used to update the local three-dimensional point cloud map according to the point cloud obtained by the robot to obtain the global three-dimensional point cloud map.

[0064] The present application also provides an electronic device, including:

[0065] at least one processor; and

[0066] a memory communicatively coupled to at least one processor; wherein

[0067] The memory stores computer instructions that can be executed by at least one processor. The computer instructions are executed by at least one processor to enable the at least one processor to perform the above method.

[0068] An embodiment of the present application also provides a computer-readable storage medium, in which a plurality of computer instructions are stored, and the plurality of computer instructions are used to enable a computer to execute the above method.

[0069] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0070] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, 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 device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0071] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0072] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0073] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0074] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for constructing a three-dimensional point cloud map, characterized in that: include: Acquire a point cloud, and construct a local three-dimensional point cloud map based on the point cloud; Determine an exploration gain at a target point according to the local three-dimensional point cloud map, and determine an optimal target point according to the exploration gain; Planning an exploration path, and driving the robot to move along the exploration path; The local three-dimensional point cloud map is updated according to the point cloud acquired by the robot to obtain a global three-dimensional point cloud map; Determining the exploration benefit at the target point according to the local three-dimensional point cloud map, and determining the best target point according to the exploration benefit, includes: Performing height filtering on the local three-dimensional point cloud map to retain point clouds within the robot's traversable height and ground range; The RANSAC algorithm is used in combination with the ground normal vector constraint fitting to obtain the ground point cloud, and the point cloud other than the ground point cloud is the non-ground point cloud; Project all point clouds onto the plane along the gravity direction. The projection height of the ground point cloud is 0, which means the ground point cloud is the area detected by the robot. The projection height of the non-ground point cloud is 1, which means the obstacle area. The projection height of other areas is -1, which means the unknown area not detected by the robot. Generate a two-dimensional grid map. estimating a target point with the largest exploration benefit in the two-dimensional grid map as the optimal target point; The estimating the target point with the maximum exploration benefit in the two-dimensional grid map as the optimal target point includes: Clustering the target points in the two-dimensional grid map by hierarchical clustering to form a plurality of clusters; Determine the comprehensive exploration benefit of the current cluster in cluster units; The cluster with the largest comprehensive exploration benefit among all clusters is selected as the target cluster, and the target point closest to the cluster center is used as the optimal target point.

2. The three-dimensional point cloud map construction method according to claim 1, characterized in that: The step of acquiring the point cloud and constructing a local three-dimensional point cloud map according to the point cloud includes: Acquire continuous frame point clouds through multi-line laser radar; According to the continuous frame point cloud, the local three-dimensional point cloud map is constructed using SLAM technology.

3. The three-dimensional point cloud map construction method according to claim 2, characterized in that: The method of constructing the local three-dimensional point cloud map using SLAM technology according to the continuous frame point cloud comprises: Using the RANSAC algorithm to perform plane fitting on the continuous frame point cloud to obtain a fitting plane; If the number of points in the fitting plane is greater than the set ratio of the number of all points in the point cloud, the fitting is stopped, and the points in the fitting plane are taken as the plane feature point cloud, and the remaining points are taken as the edge feature point cloud; Determine the smoothness of the continuous frame point clouds, use several plane feature point clouds with the highest smoothness as surface features, and use several edge feature point clouds with the lowest smoothness as edge features; The local three-dimensional point cloud map is constructed using the surface features and edge features.

4. The three-dimensional point cloud map construction method according to claim 3, characterized in that: After acquiring the continuous frame point cloud, the continuous frame point cloud is also registered. The method for registering the continuous frame point cloud includes: The T(t) matrix of the previous frame point cloud is taken as the initial solution into the edge feature {P t E , P t+1 E } set, control the z direction and rotation parameters [roll, pitch] of the multi-line laser radar in the spatial coordinate system to remain unchanged, determine the x, y and heading angle yaw of the multi-line laser radar in the spatial coordinate system.

5. The three-dimensional point cloud map construction method according to claim 2, characterized in that: After obtaining the local three-dimensional point cloud map, the local three-dimensional point cloud map is also updated, and the updating method includes: Determine the relative posture transformation relationship [R, t] through the continuous frame point cloud; The current frame point cloud is projected to the global coordinate system through the relative pose transformation relationship [R, t], inserted into the local three-dimensional point cloud map, and voxelized and downsampled, and the process is repeated to update the local three-dimensional point cloud map.

6. The three-dimensional point cloud map construction method according to claim 1, characterized in that: The planning of the exploration path and driving the robot to move along the exploration path includes: Based on the current position of the robot and the position of the optimal target point, use A The algorithm searches for a global main path by which the robot can reach the optimal target point; Based on the global main path, a local candidate path set is generated in combination with a cubic spline curve of the robot chassis dynamics, and the generated local candidate path set requires the robot to avoid collision with obstacles; Determine the probability that the global main path and the local candidate paths in the local candidate path set can reach the optimal target point, and select the path with the maximum probability as the exploration path; The robot is driven to move along the exploration path.

7. The three-dimensional point cloud map construction method according to claim 1, characterized in that: The updating of the local three-dimensional point cloud map according to the point cloud acquired by the robot to obtain a global three-dimensional point cloud map includes: In the process of the robot moving to the optimal target point, updating the global three-dimensional point cloud map and boundary point information of the scene; After the robot reaches the optimal target point, selecting the next optimal target point and navigating; The process is executed in a loop until there are no boundary points in the global three-dimensional point cloud map, thereby obtaining the updated global three-dimensional point cloud map.

8. A system using the three-dimensional point cloud map construction method according to any one of claims 1 to 7, characterized in that: include: A map construction module, used to obtain a point cloud and construct a local three-dimensional point cloud map based on the point cloud; A target determination module, used to determine the exploration benefit at the target point according to the local three-dimensional point cloud map, and determine the best target point according to the exploration benefit; A target exploration module, used for planning an exploration path and driving the robot to move along the exploration path; The map updating module is used to update the local three-dimensional point cloud map according to the point cloud acquired by the robot to obtain a global three-dimensional point cloud map.

Images