A global positioning method, system and device for laser radar in a scene

Abstract

The present invention discloses a global positioning method, system and equipment for a laser radar in a scene, and relates to the field of laser radar positioning and navigation technology. The present invention comprises: inputting local point cloud information at different positions in the scene into a local feature determination model respectively to obtain local features at different positions in the scene; constructing a scene feature database according to the local features at different positions in the scene; obtaining real-time local point cloud information collected by the laser radar at the current position at the current moment; inputting the real-time local point cloud information into the local feature determination model to obtain real-time local features; determining the position corresponding to the local feature with the greatest similarity to the real-time local feature in the scene feature database as the current position of the laser radar at the current moment. The present invention can improve the accuracy and efficiency of laser radar positioning in a scene by constructing a local feature determination model, a scene point cloud basic database and a scene feature database.

Description

Technical Field

[0001] The present invention relates to the field of laser radar positioning and navigation technology, and in particular to a global positioning method, system and equipment for a laser radar in a scene. Background Art

[0002] Scene recognition technology aims to use environmental perception data to retrieve and match scenes in the established map database, and quickly determine the current position in the global map. It is a key technology to ensure the continuity and stability of positioning and navigation in satellite signal interference or denial environments such as urban forests, indoors and underground. As the core sensor of the environmental perception system, LiDAR has the characteristics of high data resolution, strong anti-interference ability, and little influence by environmental factors. Research on scene recognition technology based on laser point cloud is of great significance for autonomous driving vehicles, robots and other intelligent mobile devices to achieve path planning and decision-making, and ensure seamless and safe navigation. When LiDAR works in complex actual scenes, there is a lot of noise that interferes with the point cloud structure. The change in point cloud density also affects the fineness of scene expression. In addition, massive data continues to flow in at a certain frame rate. How to improve the robustness of scene recognition under the influence of factors such as point cloud noise and density changes, while taking into account the real-time operation of the algorithm, is a key scientific problem that needs to be solved urgently.

[0003] Since the manually designed point cloud feature description is easily affected by factors such as point cloud sparsity and non-structure, existing methods mostly use deep learning technology to solve the point cloud scene recognition problem. Existing methods further improve the scene recognition positioning accuracy by integrating regional features and attention structures. However, when applied to actual scenes, the existing methods still have the following problems: (1) Poor real-time positioning and difficult model deployment: Existing methods improve scene recognition accuracy by designing complex network structures, ignoring the impact of network parameters on positioning timeliness, and due to the large number of parameters, it is difficult to deploy to vehicle-mounted or airborne platforms with low actual computing power; (2) Low positioning robustness in actual complex environments: Factors such as actual scene point cloud noise and density changes will seriously affect the feature perception performance of existing methods, resulting in poor positioning continuity and difficulty in meeting the robust positioning requirements of actual scenes. Summary of the invention

[0004] The purpose of the present invention is to provide a global positioning method, system and device for a laser radar in a scene, which can improve the accuracy and efficiency of real-time positioning of the laser radar in the scene.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A global positioning method of a laser radar in a scene, comprising:

[0007] Constructing a local feature determination model;

[0008] Constructing a scene point cloud basic database; the scene point cloud basic database includes local point cloud information at different locations in the scene;

[0009] Inputting local point cloud information at different positions in the scene into the local feature determination model respectively to obtain local features at different positions in the scene;

[0010] Construct a scene feature database based on local features at different locations in the scene;

[0011] Get the real-time local point cloud information collected by the laser radar at the current time and location;

[0012] Inputting the real-time local point cloud information into a local feature determination model to obtain real-time local features;

[0013] The position corresponding to the local feature with the greatest similarity to the real-time local feature in the scene feature database is determined as the current position of the laser radar at the current moment.

[0014] Optionally, the local feature determination model includes: a sampling and clustering layer, a basic feature perception layer, a local residual coding layer, an attention interaction layer, a global residual coding layer, and a generalized average pooling layer connected in sequence.

[0015] Optionally, the basic feature perception layer includes a first 2D convolutional layer, a first batch of normalization layers and a first ReLU activation layer connected in sequence.

[0016] Optionally, the local residual coding layer includes: a second 2D convolution layer, a second batch normalization layer, a second ReLU activation layer, a third 2D convolution layer, a third batch normalization layer, a third ReLU activation layer and a first maximum pooling layer connected in sequence;

[0017] Among them, the input end of the third ReLU activation layer is also connected to the output end of the basic feature perception layer.

[0018] Optionally, the global residual coding layer includes: a first 1D convolution layer, a fourth batch normalization layer, a fourth ReLU activation layer, a second 1D convolution layer, a fifth batch normalization layer, a fifth ReLU activation layer, and a second maximum pooling layer connected in sequence;

[0019] Among them, the input end of the fifth ReLU activation layer is also connected to the output end of the attention interaction layer.

[0020] Optionally, the real-time local point cloud information is input into a local feature determination model to obtain real-time local features, including:

[0021] The real-time local point cloud information is input into the sampling and clustering layer to obtain the real-time regional bottom-level features; the sampling and clustering layer uses a random sampling method to select a first preset number of points in the real-time local point cloud information as seed points; with the seed point position as the center, a K-neighborhood clustering method is used to obtain a second preset number of neighborhood points within a preset range of the seed point, and all the neighborhood points corresponding to each seed point are determined as regional bottom-level features;

[0022] Inputting the underlying features of the region into the basic feature perception layer to obtain the basic features of the region;

[0023] Inputting the regional basic features into the local residual coding layer to obtain regional fine features;

[0024] Inputting the regional fine features into the attention interaction layer to obtain regional correction features;

[0025] Inputting the region correction features into a global residual coding layer to obtain real-time region features;

[0026] Update the first preset number and the second preset number, and return to the step of "inputting the real-time local point cloud information into the sampling and clustering layer to obtain the real-time regional bottom layer features" until the preset number of iterations is reached to obtain multiple real-time regional features;

[0027] Multiple real-time regional features are input into the generalized average pooling layer to obtain point cloud global features.

[0028] A global positioning system of a laser radar in a scene, comprising:

[0029] A local feature determination model building module, used to build a local feature determination model;

[0030] A scene point cloud basic database construction module is used to construct a scene point cloud basic database; the scene point cloud basic database includes local point cloud information at different locations in the scene;

[0031] A local feature determination module is used to input local point cloud information at different positions in the scene into the local feature determination model to obtain local features at different positions in the scene;

[0032] A scene feature database construction module is used to construct a scene feature database based on local features at different locations in the scene;

[0033] A real-time local point cloud information acquisition module is used to obtain the real-time local point cloud information collected by the laser radar at the current time and current position;

[0034] A real-time local feature determination module, used for inputting the real-time local point cloud information into a local feature determination model to obtain real-time local features;

[0035] The laser radar real-time position determination module is used to determine the position corresponding to the local feature in the scene feature database that has the greatest similarity to the real-time local feature as the current position of the laser radar at the current moment.

[0036] An electronic device comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the global positioning method of a laser radar within a scene.

[0037] Optionally, the memory is a readable storage medium.

[0038] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0039] The present invention provides a method, system and device for global positioning of a laser radar in a scene, including: constructing a local feature determination model; constructing a scene point cloud basic database; the scene point cloud basic database includes local point cloud information at different positions in the scene; inputting the local point cloud information at different positions in the scene into the local feature determination model respectively to obtain local features at different positions in the scene; constructing a scene feature database according to the local features at different positions in the scene; obtaining real-time local point cloud information collected by the laser radar at the current position at the current moment; inputting the real-time local point cloud information into the local feature determination model to obtain real-time local features; determining the position corresponding to the local feature with the greatest similarity to the real-time local feature in the scene feature database as the current position of the laser radar at the current moment. The present invention can improve the accuracy and efficiency of real-time positioning of the laser radar in the scene by constructing a local feature determination model, a scene point cloud basic database and a scene feature database. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0041] Figure 1 This is a flow chart of the global positioning method of the laser radar in the scene in Example 1 of the present invention;

[0042] Figure 2 This is a schematic diagram of the local feature determination model structure in Example 1 of the present invention;

[0043] Figure 3 Schematic diagram of the structure of a local residual coding layer and a global residual coding layer in Embodiment 1 of the present invention;

[0044] Figure 4 Schematic diagram of the attention interaction layer structure in Example 1 of the present invention;

[0045] Figure 5 This is a flow chart of positioning search in Example 1 of the present invention;

[0046] Figure 6 A comparison chart of scene matching accuracy and model parameter quantity between the global positioning method in Example 1 of the present invention and the existing scene recognition positioning method;

[0047] Figure 7 This is a diagram showing the actual scene positioning results in Example 1 of the present invention;

[0048] Figure 8 This is a diagram showing the positioning results of strong noise and low-density scenes in Example 1 of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The purpose of the present invention is to provide a global positioning method, system and device for a laser radar in a scene, which can improve the accuracy and efficiency of real-time positioning of the laser radar in the scene.

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] Example 1

[0053] like Figure 1 As shown, this embodiment provides a global positioning method of a laser radar in a scene (referred to as LWR-Net), including:

[0054] Step 101: Construct a local feature determination model. Figure 2, Optionally, the local feature determination model includes: a sampling and clustering layer, a basic feature perception layer, a local residual coding layer, an attention interaction layer, a global residual coding layer, and a generalized average pooling layer connected in sequence. The basic feature perception layer includes a first 2D convolution layer, a first batch of standardization layers, and a first ReLU activation layer connected in sequence. The local residual coding layer includes: a second 2D convolution layer, a second batch of standardization layers, a second ReLU activation layer, a third 2D convolution layer, a third batch of standardization layers, a third ReLU activation layer, and a first maximum pooling layer connected in sequence; the input end of the third ReLU activation layer is also connected to the output end of the basic feature perception layer. The global residual coding layer includes: a first 1D convolution layer, a fourth batch of standardization layers, a fourth ReLU activation layer, a second 1D convolution layer, a fifth batch of standardization layers, a fifth ReLU activation layer, and a second maximum pooling layer connected in sequence; wherein the input end of the fifth ReLU activation layer is also connected to the output end of the attention interaction layer (that is, the output end of the attention interaction layer is connected to the input end of the xth ReLU activation layer in the form of a residual).

[0055] Step 102: Construct a scene point cloud basic database; the scene point cloud basic database includes local point cloud information at different locations in the scene. Obtain point cloud information at different locations in the scene and construct a scene point cloud basic database. The scene point cloud basic database is mainly composed of local area point clouds at different locations of the scene and their corresponding sampling locations.

[0056] Step 103: inputting the local point cloud information at different positions in the scene into the local feature determination model respectively to obtain the local features at different positions in the scene.

[0057] This step receives a single local scene point cloud as input. Suppose the input local scene point cloud is Where N is the number of points in the local scene point cloud, 3 represents the three-dimensional coordinate information of the input point cloud, and the global feature description of the local scene point cloud in 256 dimensions is output for scene matching and positioning. The specific technical solution is as follows:

[0058] Step 103-1: Sampling and clustering layer: For the input point cloud P, use the random sampling method to obtain 1024 random point positions in the input scene point cloud as seed points. Let the seed points after sampling be Then, with the seed point as the center, the K-neighborhood clustering method is used to obtain 32 neighboring points around the seed point as the regional underlying features of the input point cloud P, denoted as

[0059] Step 103-2: Basic feature perception layer: based on the local area underlying feature P k1As input, the feature dimension is first increased to 64 through a 2D convolution layer. The convolution kernel size used in the convolution layer is 1*1. Then, the regional basic features are obtained through a batch normalization layer and a ReLU activation layer.

[0060] Step 103-3: Local residual coding layer: based on regional basic features F B1 The local residual coding layer is specifically structured as follows: Figure 3 As shown, F B1 After passing through the 2D convolution layer, batch normalization layer, ReLU activation layer, 2D convolution layer, batch normalization layer, it is added to the original feature in the feature dimension, and finally input into the ReLU activation layer, and the maximum pooling layer is used to reduce the feature dimension at the neighborhood scale to obtain the regional fine features, which are recorded as

[0061] Step 103-4: Attention interaction layer: using regional fine features F R1 As input, the main structure is as follows Figure 4 As shown, F R1 First, three convolution layers with a convolution kernel size of 1*1 are used to generate Query, Key, and Value features. The Query, Key, and Value features do not change the input feature dimension. The convolution layer used to generate Query and Key features is a grouped convolution structure with the number of groups set to 2. The feature dimension is divided using the set number of groups, and convolution operations are performed within each group to reduce the number of parameters in the attention interaction layer. The Query and Key features are obtained by matrix multiplication to obtain the attention weight feature matrix, which is recorded as Express the correlation between the information of each local area of ​​the Key and Query features. W1 and F R1 The attention residual features are obtained by matrix multiplication and combined with F R1 Add in the feature dimension to obtain the regional correction feature

[0062] Step 103-5: Global residual coding layer 1: regional correction feature F A1 As input, its specific structure is as follows Figure 3 As shown. A1 After passing through the 1D convolution layer, batch normalization layer, ReLU activation layer, 1D convolution layer, batch normalization layer, it is added to the original feature in the feature dimension, and finally input into the ReLU activation layer, and the maximum pooling layer is used to reduce the feature dimension at the neighborhood scale to obtain the final regional feature, which is recorded as

[0063] Step 103-6: Iterate steps 103-1 to 103-5 to further expand the receptive field and feature dimension based on the current local area features. The number of points sampled in the local area of ​​the sampling and clustering layer is 256, the number of points in the neighborhood clustering is 16, the feature enhancement dimension of the basic feature perception layer is 256, and the output feature is recorded as

[0064] Step 103-7: Feature F G2 As input, the global features of the scene point cloud with 256 dimensions are obtained through the generalized average pooling layer.

[0065] The local scene point clouds in the scene point cloud database are traversed as the input of the local feature determination model, and the output local scene features are used to build a scene feature database. Among them, the main components of the scene feature database are the global 256-dimensional features of local scenes at different locations and the sampling positions of the scene point clouds.

[0066] Step 104: Construct a scene feature database based on local features at different locations in the scene.

[0067] Step 105: Obtain real-time local point cloud information collected by the laser radar at the current time and position.

[0068] Step 106: Input the real-time local point cloud information into the local feature determination model to obtain the real-time local features.

[0069] Step 106 includes:

[0070] Step 1061: input the real-time local point cloud information into the sampling and clustering layer to obtain the real-time regional underlying features; the sampling and clustering layer uses a random sampling method to use a first preset number of points in the real-time local point cloud information as seed points; with the seed point position as the center, use the K-neighborhood clustering method to obtain a second preset number of neighborhood points within the preset range of the seed point, and determine all the neighborhood points corresponding to each seed point as the regional underlying features.

[0071] Step 1062: Input the underlying features of the region into the basic feature perception layer to obtain the basic features of the region.

[0072] Step 1063: Input the regional basic features into the local residual coding layer to obtain the regional fine features.

[0073] Step 1064: Input the regional fine features into the attention interaction layer to obtain the regional correction features.

[0074] Step 1065: Input the region correction features into the global residual coding layer to obtain real-time region features.

[0075] Step 1066: Update the first preset number and the second preset number, and return to the step of "inputting the real-time local point cloud information into the sampling and clustering layer to obtain the real-time regional underlying features" until the preset number of iterations is reached to obtain multiple real-time regional features.

[0076] Step 1067: Input multiple real-time regional features into the generalized average pooling layer to obtain real-time local features.

[0077] Step 107: Determine the position corresponding to the local feature in the scene feature database that has the greatest similarity to the real-time local feature as the current position of the laser radar at the current moment. When the target device issues a positioning request, the laser radar and other sensors carried by the device are used to obtain the local scene point cloud. After the global 256-dimensional features of the scene are obtained through the local feature determination model, the scene feature database is searched and matched based on the feature distance. When the corresponding scene is retrieved, the sampling position of the scene is retrieved and provided to the target device as the current global positioning result. The retrieval and matching positioning process is as follows: Figure 5 shown.

[0078] like Figure 6-Figure 8 As shown in the figure, PointNetVLAD is a point cloud local aggregation feature description network; LPD-Net is a point cloud recognition network for large-scale scenes; PCAN is a point cloud scene recognition network integrating contextual attention; SOE-Net is a point cloud scene recognition network integrating self-attention and direction encoding; MinkLoc3D: a point cloud scene recognition network based on sparse voxel convolution; EPC-Net is a fast point cloud scene recognition network; NDT-Tranformer is a point cloud scene recognition network based on normal distribution transformation; PPT-Net is a point cloud scene recognition network based on pyramid attention; LWR-Net is an ultra-lightweight point cloud scene recognition network; the scene recognition and positioning method proposed in the present invention is superior to the existing The method has three major improvements: (1) Strong robustness in complex scenes such as strong noise and low density: Compared with existing methods, the method proposed in the present invention has better robustness in scenes with strong noise and low-density point cloud. Affected by complex scenes, the scene recognition and positioning accuracy of existing methods is greatly attenuated. The method proposed in the present invention can maintain high recognition accuracy and positioning robustness; (2) Ultra-lightweight model size: The model parameters of existing scene recognition methods for lidar global positioning are relatively large. The scene recognition and positioning network parameters proposed in the present invention are only 300,000, which is more than twice the size of existing methods; (3) Strong scene generalization ability: Compared with existing methods, the method proposed in the present invention can better adapt to new scenes and maintain positioning robustness under scene environment and style changes.

[0079] Example 2

[0080] In order to execute the method corresponding to the above-mentioned embodiment 1 to achieve corresponding functions and technical effects, a global positioning system of a laser radar in a scene is provided below, including:

[0081] The local feature determination model building module is used to build a local feature determination model.

[0082] The scene point cloud basic database construction module is used to construct the scene point cloud basic database. The scene point cloud basic database includes local point cloud information at different locations in the scene.

[0083] The local feature determination module is used to input the local point cloud information at different positions in the scene into the local feature determination model to obtain the local features at different positions in the scene.

[0084] The scene feature database construction module is used to construct a scene feature database based on local features at different locations in the scene.

[0085] The real-time local point cloud information acquisition module is used to obtain the real-time local point cloud information collected by the laser radar at the current moment and current position.

[0086] The real-time local feature determination module is used to input the real-time local point cloud information into the local feature determination model to obtain the real-time local features.

[0087] The laser radar real-time position determination module is used to determine the position corresponding to the local feature in the scene feature database that has the greatest similarity to the real-time local feature as the current position of the laser radar at the current moment.

[0088] Example 3

[0089] This embodiment provides an electronic device, including a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the global positioning method of a laser radar in a scene described in Example 1.

[0090] Among them, the memory is a readable storage medium.

[0091] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0092] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A global positioning method of laser radar in a scene, characterized in that: include: Constructing a local feature determination model; Constructing a scene point cloud basic database; the scene point cloud basic database includes local point cloud information at different locations in the scene; Inputting local point cloud information at different positions in the scene into the local feature determination model respectively to obtain local features at different positions in the scene; The local feature determination model includes: a sampling and clustering layer, a basic feature perception layer, a local residual coding layer, an attention interaction layer, a global residual coding layer, and a generalized average pooling layer connected in sequence; the local residual coding layer includes: a second 2D convolution layer, a second batch normalization layer, a second ReLU activation layer, a third 2D convolution layer, a third batch normalization layer, a third ReLU activation layer, and a first maximum pooling layer connected in sequence; wherein the input end of the third ReLU activation layer is also connected to the output end of the basic feature perception layer; the global residual coding layer includes: a first 1D convolution layer, a fourth batch normalization layer, a fourth ReLU activation layer, a second 1D convolution layer, a fifth batch normalization layer, a fifth ReLU activation layer, and a second maximum pooling layer connected in sequence; wherein the input end of the fifth ReLU activation layer is also connected to the output end of the attention interaction layer; Construct a scene feature database based on local features at different locations in the scene; Get the real-time local point cloud information collected by the laser radar at the current time and location; Inputting the real-time local point cloud information into a local feature determination model to obtain real-time local features; The position corresponding to the local feature with the greatest similarity to the real-time local feature in the scene feature database is determined as the current position of the laser radar at the current moment.

2. The global positioning method of laser radar in a scene according to claim 1, characterized in that: The basic feature perception layer includes a first 2D convolutional layer, a first batch of normalization layers and a first ReLU activation layer which are connected in sequence.

3. The global positioning method of laser radar in a scene according to claim 1, characterized in that: Inputting the real-time local point cloud information into a local feature determination model to obtain real-time local features includes: The real-time local point cloud information is input into the sampling and clustering layer to obtain the real-time regional bottom-level features; the sampling and clustering layer uses a random sampling method to select a first preset number of points in the real-time local point cloud information as seed points; with the seed point position as the center, a K-neighborhood clustering method is used to obtain a second preset number of neighborhood points within a preset range of the seed point, and all the neighborhood points corresponding to each seed point are determined as regional bottom-level features; Inputting the underlying features of the region into the basic feature perception layer to obtain the basic features of the region; Inputting the regional basic features into the local residual coding layer to obtain regional fine features; Inputting the regional fine features into the attention interaction layer to obtain regional correction features; Inputting the region correction features into a global residual coding layer to obtain real-time region features; Update the first preset number and the second preset number, and return to the step of "inputting the real-time local point cloud information into the sampling and clustering layer to obtain the real-time regional bottom layer features" until the preset number of iterations is reached to obtain multiple real-time regional features; Multiple real-time regional features are input into the generalized average pooling layer to obtain point cloud global features.

4. A global positioning system of laser radar in a scene, characterized in that: include: A local feature determination model building module, used to build a local feature determination model; A scene point cloud basic database construction module is used to construct a scene point cloud basic database; the scene point cloud basic database includes local point cloud information at different locations in the scene; A local feature determination module is used to input local point cloud information at different positions in the scene into the local feature determination model to obtain local features at different positions in the scene; The local feature determination model includes: a sampling and clustering layer, a basic feature perception layer, a local residual coding layer, an attention interaction layer, a global residual coding layer, and a generalized average pooling layer connected in sequence; the local residual coding layer includes: a second 2D convolution layer, a second batch normalization layer, a second ReLU activation layer, a third 2D convolution layer, a third batch normalization layer, a third ReLU activation layer, and a first maximum pooling layer connected in sequence; wherein the input end of the third ReLU activation layer is also connected to the output end of the basic feature perception layer; the global residual coding layer includes: a first 1D convolution layer, a fourth batch normalization layer, a fourth ReLU activation layer, a second 1D convolution layer, a fifth batch normalization layer, a fifth ReLU activation layer, and a second maximum pooling layer connected in sequence; wherein the input end of the fifth ReLU activation layer is also connected to the output end of the attention interaction layer; A scene feature database construction module is used to construct a scene feature database based on local features at different locations in the scene; A real-time local point cloud information acquisition module is used to obtain the real-time local point cloud information collected by the laser radar at the current time and current position; A real-time local feature determination module, used for inputting the real-time local point cloud information into a local feature determination model to obtain real-time local features; The laser radar real-time position determination module is used to determine the position corresponding to the local feature in the scene feature database that has the greatest similarity to the real-time local feature as the current position of the laser radar at the current moment.

5. An electronic device, characterized in that: It includes a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute a global positioning method of a laser radar in a scene as described in any one of claims 1 to 3.

6. An electronic device according to claim 5, characterized in that: The memory is a readable storage medium.

Images