A method for dynamic segmentation and loading of a point cloud map based on a topological structure

Abstract

The present invention provides a method for dynamic segmentation and loading of point clouds based on a topological structure, including obtaining a point cloud map of a scene and establishing a topological map according to the topological structure of the scene; for segmentation: defining the attributes of topological nodes, defining a cuboid with a topological node as the three-dimensional center, and calculating all the point cloud points within the cuboid; preprocessing the point cloud points to generate a sub-point cloud map corresponding to the topological node; traversing the topological nodes to obtain a set of sub-point cloud maps; storing the sub-point cloud maps corresponding to all topological nodes, and at the same time storing a topological structure description file. According to the topological structure of the scene, the present invention segments the point cloud map of the scene according to topological nodes, and at the same time loads the point cloud map in real time according to requirements, realizes the real-time update of the point cloud map, and improves the reuse rate and update efficiency of the map.

Description

Technical Field

[0001] The present invention belongs to the field of point cloud processing, and particularly relates to a method for dynamically segmenting and loading a point cloud map based on a topological structure. Background Art

[0002] Applying driverless technology to large scenarios and using a global point cloud map for real-time positioning will greatly improve the stability and global unity of the positioning of autonomous vehicles, and also help simplify the complexity of planning and control. For large scenarios, especially for driverless in large-scale similar scenarios such as long tunnels, the implementation of robust positioning can significantly expand the application scope of driverless, improve the degree of industrial integration, and promote the progress of driverless technology.

[0003] Generally, a point cloud refers to a massive set of points representing the surface characteristics of an object. The point cloud obtained according to the laser measurement principle includes three-dimensional coordinates (XYZ) and laser reflection intensity (Intensity). For large-scale scene positioning, especially for indoor scenes of more than ten kilometers, such as tunnels and underground mines, a point cloud map of the entire scene is required, which will be a very large amount of data. It is not only difficult for computing devices to load and import, but also extremely difficult for algorithm calculations. However, for a single vehicle, the autonomous driving positioning of the vehicle is usually within a small range, and this range is generally related to the detection sensor. Therefore, most of the point cloud map is actually "unnecessary" temporarily, which means it does not need to participate in the calculation and does not need to be loaded and imported into the computer memory.

[0004] Up to now, in the existing technologies, the methods for segmenting and loading a point cloud map generally divide the point cloud map according to general rules, and the obtained point cloud maps are often sub-maps of the same shape, which are difficult to relate to the actual scene structure; moreover, when importing such maps, due to the requirements of real-time positioning, multiple sub-maps generally need to be imported, and there is also some "waste" of maps; in addition, in actual use, for some similar but intermittently changing scenes, real-time maintenance cannot be achieved.

[0005] Therefore, in view of the above requirements and problems, the present invention proposes a method for dynamically segmenting and loading a point cloud map based on a topological structure, which effectively solves the above problems. Summary of the Invention

[0006] To this end, the present invention provides a method for dynamically segmenting and loading a point cloud based on a topological structure.

[0007] The method for dynamically segmenting and loading a point cloud based on a topological structure of the present invention includes:

[0008] Obtaining a point cloud map of a scene and establishing a topological map according to the topological structure of the scene;

[0009] For segmentation:

[0010] Define the attributes of the topological nodes, define a cuboid with the topological node as the three-dimensional center, and calculate all the point cloud points within the cuboid;

[0011] Preprocess the point cloud points to generate a sub-point cloud map corresponding to the topological node;

[0012] Traverse the topological nodes to obtain a set of sub-point cloud maps;

[0013] Store the sub-point cloud maps corresponding to all topological nodes, and at the same time store the topological structure description file.

[0014] Furthermore, when used for loading, it includes the following steps:

[0015] Read the topological structure description file to generate a topological map;

[0016] According to the index position, calculate the topological region where this position is located, read the topological nodes in this topological region, and obtain the attribute information of the topological nodes; the index position is the position specified manually or required by device positioning;

[0017] Judge whether the attribute information needs to be changed,

[0018] If it needs to be changed, generate a sub-point cloud map according to the topological node after the attribute information is changed, and store the updated topological structure description file;

[0019] If it does not need to be changed, calculate the adjacent topological node information according to the index position, and at the same time sort out all the topological nodes required, and read the corresponding sub-point cloud maps from the specified path;

[0020] Merge the read sub-point cloud maps into the point cloud map to be published, and then publish this point cloud map.

[0021] Furthermore, the attributes of the topological nodes include identity ID, position, a corner point of the cross-section, the diagonal point of the cross-section, the cross-section length, the sampling method, and the sampling rate.

[0022] Furthermore, among the attributes of the topological node, the identity ID is unique.

[0023] Furthermore, when storing the sub-point cloud maps corresponding to all topological nodes, store them separately and uniquely with the identity ID of the topological node.

[0024] The present invention also provides a point cloud dynamic segmentation and loading system based on a topological structure, including:

[0025] An offline segmentation module that outputs a sub-point cloud map and a topological structure description file after inputting a point cloud map and a topological structure description file;

[0026] A dynamic loading module, after inputting the sub-point cloud map and the topological structure description file, changes the point cloud map relationship according to external requirements and publishes the point cloud map.

[0027] The present invention also provides an electronic device, including: a memory, a processor, and the topological structure-based point cloud dynamic segmentation and loading system as described in the present invention; the processor is connected to the topological structure-based point cloud dynamic segmentation and loading system and the memory through a bus; the memory is used to store computer execution instructions; the processor is used to execute the computer execution instructions stored in the memory.

[0028] The present invention also provides a storage medium, a readable storage medium, and a computer program, and the computer program is used to control the topological structure-based point cloud dynamic segmentation and loading method as described in the present invention.

[0029] The above technical solution of the present invention has the following advantages compared with the prior art:

[0030] The topological structure-based point cloud dynamic segmentation and loading method provided by the present invention can directly set a topological structure description file according to actual environmental characteristics, project business requirements, etc. According to this file, the large-scale scene point cloud map can be separated offline, improving the utilization efficiency of the sub-map and reducing the probability of secondary repeated segmentation; in addition, after importing the sub-map and the topological structure relationship description file, the local sub-map information can be updated according to actual needs, and the local sub-map information can be published, improving the map update efficiency and the publishing efficiency, and helping to improve the stability and rapidity of other systems (such as the positioning system). BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the overall system structure of the present invention;

[0032] Figure 2 It is a schematic diagram of the three-dimensional space of a cuboid included in a topological node;

[0033] Figure 3 It is a program flow chart of the offline segmentation module;

[0034] Figure 4 It is a program flow chart of the dynamic loading module;

[0035] Figure 5 It is a point cloud map of the ground part of a certain underground mine scene;

[0036] Figure 6 It is a conceptual diagram of the topological structure in the underground mine scene;

[0037] Figure 7 It is a partial point cloud sub-map after segmentation;

[0038] Figure 8 It is a local map automatically loaded by the dynamic loading module in the ground scene;

[0039] Figure 9 It is a local map automatically loaded by the dynamic loading module in the underground scene;

[0040] Figure 10 It is a local map automatically loaded after setting the topological structure parameters for the dynamic loading module. Specific implementation manners

[0041] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0042] Taking a certain well mining scene as an example, the point cloud dynamic loading and segmentation system based on the topological structure in this embodiment uses a Suteng 32-line mechanical scanning lidar rslidar32, a Dai's combined inertial navigation device IFS2000, a wheel speedometer and other sensors to conduct a positioning experiment based on the point cloud map.

[0043] First, all point cloud maps in this scene are obtained. Through preliminary analysis, it can be seen that generally the ground scene is open, and the lidar can sense the surrounding area of about 70 meters, while the underground is narrow and generally can only sense the range of about 50 meters before and after in the forward direction. Therefore, this scene can be divided into two types, that is, there are two types of topological node types.

[0044] First, the attributes of the topological nodes need to be set. It should be noted that the attributes of the topological nodes vary according to the working scene and actual requirements. In this embodiment, the ground scene and the underground scene are taken as examples.

[0045] For the ground scene type, some attributes of the topological nodes can be set as follows:

[0046]

[0047] Among them, the corner point coordinates are set as the offset relative to the node position, the ID value is automatically incremented according to the quantity, and the position is obtained according to the actual position of the point cloud map.

[0048] For the underground scene type, some attributes of the topological nodes can be set as follows:

[0049]

[0050] Among them, due to the single underground characteristics, the downsampling rate should be slightly larger than that of the ground scene, and the point cloud density can be appropriately reduced to improve the point cloud retrieval speed.

[0051] After setting, load the point cloud map and the topological structure description file through the offline segmentation module. The calculation steps are as follows:

[0052] Load the point cloud map, load the topological relationship in the topological map, and generate the topological map.

[0053] Traverse the topological nodes to obtain the attribute values of the topological nodes. Assume the position of the topological node is v i (x i , y i , z i ). Taking this position of the topological node as the center point, obtain an array:

[0054] [x_min, y_min, z_min, x_max, y_max, z_max]

[0055] This array is used to describe the cuboid centered on this topological node.

[0056] If this topological node is on the ground, according to the topological node attributes set on the ground, then this array is:

[0057] [-100, -100, -100, 100, 100, 100];

[0058] If this topological node is underground, according to the topological node attributes set underground, then this array is:

[0059] [-10, -200, -10, 10, 200, 10];

[0060] Then, calculate the point cloud points within the spatial range through the CropBoxfilter filter.

[0061] Then, downsample the point cloud through the VoxelGridfilter filter.

[0062] Save the downsampled point cloud as a sub-point cloud map file prefixed with the ID number, and at the same time update the topological structure description file.

[0063] Continue to index each subsequent topological node until all topological nodes are traversed.

[0064] Finally, save the topological structure description file.

[0065] Next, using the topological structure description file generated in this embodiment, map loading can be performed. The loading process during loading is as follows:

[0066] First, read the topology structure description file to generate the corresponding topology map.

[0067] According to the initial position p0(x0, y0, z0) provided by the positioning system, calculate the distance to each topology node v i as follows:

[0068] ||p0 - v i ||2 < σ, v i ∈ V

[0069] The calculated topology node ID is 1, that is, v1.

[0070] Read the topology node v1 of this topology region from the topology map to obtain the relevant attribute information.

[0071] Since it is the initial state and the system has not made information changes, directly calculate the sensing range of the current lidar sensor. According to this sensing range, calculate the set of neighboring topology nodes.

[0072] ||v i - v1||2 < σ, v i ∈ V & i ≠ 1

[0073] The calculated set of topology node IDs is {2, 3, 4, 5, 6}, that is, {v1, v2, v3, v4, v5, v6}.

[0074] Directly read the point cloud sub - maps corresponding to the set of topology nodes {v1, v2, v3, v4, v5, v6}.

[0075] Finally, merge the sub - point cloud maps and then publish the map.

[0076] Return the calculation of the distance between the position provided by the positioning system and each topology node. When it is found that the position provided by the positioning system is moving away from the current topology region and the ratio of the map area sensed by the lidar to the current map area is less than a certain threshold, start local map update.

[0077] Repeat the above loading process, then return the calculation of the distance between the position provided by the positioning system and each topology node, and continue to monitor the position provided by the positioning system.

[0078] During a certain update process, if it is found that in a similar scenario (in the underground environment), the actual scenario is longer than the map description, then the parameter of "section length (y)" can be modified at this time.

[0079] According to this change information, the system will change the relevant topological relationships and attribute values. According to a certain proportional relationship and ensure that the ground Figure 1In the case of consistency, perform the "copy and paste" operation on the sub-map corresponding to the topological node according to the corresponding ratio to complete part of the update of the sub-map.

[0080] Then, save the changed sub-map, and at the same time save the updated topological description file; then, calculate the set of adjacent topological nodes, and continue to execute the loading process in this embodiment.

[0081] According to the actual environmental characteristics, project business requirements, etc., a topological structure description file can be directly set. According to this file, the large-scale point cloud map can be separated offline, improving the utilization efficiency of the sub-map and reducing the probability of secondary repeated segmentation; in addition, after importing the sub-map and the topological structure relationship description file, the local sub-map information can be updated according to actual needs, and the local sub-map information can be published, improving the map update efficiency and publishing efficiency, and helping to improve the stability and speed of other systems (such as the positioning system).

[0082] This embodiment also provides a point cloud dynamic segmentation and loading system based on a topological structure, including:

[0083] An offline segmentation module that outputs a sub-point cloud map and a topological structure description file after inputting a point cloud map and a topological structure description file;

[0084] A dynamic loading module that changes the point cloud map relationship according to external requirements and publishes the point cloud map after inputting the sub-point cloud map and the topological structure description file.

[0085] This embodiment also provides an electronic device, which includes: a memory, a processor, and the point cloud dynamic segmentation and loading system based on a topological structure as described in this embodiment; the processor is connected to the point cloud dynamic segmentation and loading system based on a topological structure and the memory through a bus; the memory is used to store computer execution instructions; the processor is used to execute the computer execution instructions stored in the memory.

[0086] This embodiment also provides a storage medium, including: a readable storage medium and a computer program, and the computer program is used to control the point cloud dynamic segmentation and loading method as described in this embodiment.

[0087] Obviously, the above embodiments are merely examples given for clear illustration, and are not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. And the obvious changes or modifications derived therefrom are still within the protection scope of the present invention.

Claims

1. A method for dynamic segmentation and loading of point clouds based on topological structure, characterized in that It includes the following steps: Obtain the point cloud map of the scene, establish a topological map according to the topological structure of the scene, and at the same time obtain the topological structure description file; When used for segmentation: Define the attributes of topological nodes, define a cuboid with the topological node as the three-dimensional center, and calculate all the point cloud points within the cuboid; Preprocess the point cloud points to generate a sub-point cloud map corresponding to the topological node, and update the topological structure description file; Traverse the topological nodes to obtain a set of sub-point cloud maps; Store the sub-point cloud maps corresponding to all topological nodes, and at the same time store the topological structure description file; When used for loading, it includes the following steps: Read the topological structure description file to generate a topological map; According to the index position, calculate the topological region where this position is located, read the topological nodes in this topological region, and obtain the attribute information of the topological nodes; the index position is the position specified manually or required by device positioning; Judge whether the attribute information needs to be changed, If it needs to be changed, generate a sub-point cloud map according to the topological node after the attribute information is changed, and store the updated topological structure description file; If it does not need to be changed, according to the index position, calculate the adjacent topological node information, and at the same time sort out all the topological nodes required, and read the corresponding sub-point cloud map from the specified path; Merge the read sub-point cloud maps into the point cloud map to be published, and then publish the point cloud map; The attributes of the topological nodes include identity ID, position, a corner point of the cross-section, the diagonal point of the cross-section, the cross-section length, the sampling method, and the sampling rate; Among the attributes of the topological nodes, the identity ID is unique; When storing the sub-point cloud maps corresponding to all topological nodes, store them separately and uniquely with the identity ID of the topological node.

2. A point cloud dynamic segmentation and loading system based on a topological structure, applying the method for point cloud dynamic segmentation and loading based on a topological structure as claimed in claim 1, characterized in that, It includes: An offline segmentation module that outputs a sub-point cloud map and a topological structure description file after inputting a point cloud map and a topological structure description file; A dynamic loading module that changes the point cloud map relationship according to external requirements and publishes the point cloud map after inputting a sub-point cloud map and a topological structure description file.

3. An electronic device, characterized in that, It includes: A memory, a processor, and a point cloud dynamic segmentation and loading system based on topological structure as described in claim 2; The processor is connected to the point cloud dynamic segmentation and loading system based on topological structure and the memory through a bus; The memory is used to store computer execution instructions; The processor is used to execute the computer execution instructions stored in the memory.

4. A storage medium, characterized in that, It includes: A readable storage medium and a computer program, and the computer program is used to implement the point cloud dynamic segmentation and loading method based on topological structure as described in claim 1.

Images