Dynamic point cloud elimination method for laser radar synchronous positioning and mapping

Abstract

The invention discloses a dynamic point cloud elimination method for laser radar synchronous positioning and mapping. The method comprises the following steps: step 1, designing a point cloud semantic segmentation network; 2, training a point cloud semantic segmentation network; step 3, carrying out semantic segmentation on the point cloud; step 4, dynamic point cloud elimination; 5, a front-end odometer is arranged; and step 6, performing back-end optimization and map construction. Compared with the prior art, the method has the advantages that the dynamic point cloud in the laser point cloud data is effectively removed by introducing the semantic segmentation technology, so that the point cloud registration precision and the point cloud map construction accuracy are improved.

Description

Technical Field

[0001] The present invention relates to vehicle positioning and mapping technology methods, and in particular to a dynamic point cloud elimination method for laser radar synchronous positioning and mapping. Background Art

[0002] In the field of autonomous driving, laser radar simultaneous localization and mapping (SLAM) technology has been widely used to locate vehicles and build high-precision maps of the surrounding environment. However, laser SLAM systems often face many challenges in dynamic mapping, such as dynamic objects (including pedestrians and moving vehicles) that interfere with point cloud registration and map construction.

[0003] In the prior art, a variety of methods have been proposed for the processing and application of lidar point cloud data, but less attention has been paid to the problem of eliminating dynamic point cloud data. For example, patent CN118429637A proposes a semantic segmentation method for point clouds at road intersections, which improves the semantic segmentation performance through an improved RandLA-Net network and a weighted cross entropy loss function, but does not specifically target the elimination of dynamic point clouds. Patent CN118447248A proposes a method for generating high-precision maps for autonomous driving based on lidar. Although it adopts an unordered point cloud semantic segmentation algorithm based on PointNet, it also does not focus on the processing of dynamic point clouds. In addition, some prior arts eliminate dynamic point clouds through geometric features and motion information, but these methods have problems such as complex parameter settings and high requirements on sensor accuracy and environmental conditions. For example, patent CN116385292A proposes an IMU-assisted LiDAR dynamic point cloud elimination method. This method collects LiDAR and IMU data, performs time synchronization and coordinate system relative space conversion parameter calibration, uses IMU information to assist in eliminating LiDAR point cloud motion distortion, and determines the motion state of point cloud clusters through a multi-level fuzzy comprehensive evaluation model, thereby eliminating dynamic point clouds. However, there are still difficulties in setting the parameters of the evaluation model, which may affect the accuracy and application scope of the model.

[0004] In summary, in terms of dynamic point cloud data removal, further research and development is still needed to improve the recognition ability of dynamic objects, optimize the efficiency and accuracy of the removal algorithm, reduce the dependence on sensor accuracy and environmental conditions, and ultimately achieve robust performance and high-precision positioning and mapping of SLAM systems in various complex dynamic environments. In addition, simplifying the parameter setting process and improving the adaptability and generalization ability of the model are also important directions for promoting the development of technology in this field. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a dynamic point cloud elimination method for simultaneous positioning and mapping of laser radar.

[0006] Technical solution: The dynamic point cloud elimination method for laser radar synchronous positioning and mapping described in the present invention comprises the following steps:

[0007] Step 1: Point cloud semantic segmentation network design: Taking into account hardware resource utilization, computing efficiency, and training complexity, the SalsaNext neural network is selected as the basic model for the semantic segmentation of lidar point cloud data. Based on the SalsaNext neural network, the lidar point cloud data semantic segmentation network SalsaNext++ is designed;

[0008] Step 2: Point cloud semantic segmentation network training: Use the point cloud dataset Semantic-KITTI to train the semantic segmentation network, extract features and classify the input point cloud data, and assign the point cloud to predefined semantic categories, such as vehicles, pedestrians, and buildings;

[0009] Step 3: Point cloud semantic segmentation: Integrate the trained SalsaNext++ semantic segmentation network into the LiDAR simultaneous positioning and mapping front-end system to process LiDAR point cloud data in real time, and identify dynamic point clouds (such as moving vehicles, pedestrians, etc.) and static point clouds (such as roads, buildings, etc.) through semantic segmentation;

[0010] Step 4: Dynamic point cloud removal: For the identified dynamic point clouds, according to the semantic labels added when assigned to the predefined semantic categories, they are removed through the point cloud removal algorithm and program, and only the static point clouds are retained for subsequent vehicle pose estimation and map construction;

[0011] Step 5: Front-end odometer: In the front-end odometer module of the LiDAR simultaneous positioning and mapping system, the static point cloud data after the dynamic point cloud is removed is used to quickly estimate the pose transformation between adjacent LiDAR data frames through a feature-based point cloud registration method, and the motion of the LiDAR is estimated for point cloud dedistortion, providing inter-frame pose transformation and dedistorted static point cloud for LiDAR simultaneous positioning and mapping back-end optimization;

[0012] Step 6: Backend optimization and map construction: The backend optimization and map construction module receives the dedistorted point cloud and inter-frame pose transformation of the front-end odometer, performs global registration on multiple frame point clouds, fuses the point cloud data into the world map for map update, and fuses the obtained global pose transformation with the inter-frame pose transformation to estimate the motion state of the lidar relative to the map.

[0013] Preferably, step 1 includes the following sub-steps:

[0014] Step 1-1: The SalsaNext network proposed in 2020 still ranks at the top of the list of semantic segmentation of point cloud in the Semantic-KITTI dataset, and it does not require high hardware storage and computing power, and its training complexity and training time are relatively low, so the SalsaNext neural network is selected as the basic model for semantic segmentation of lidar point cloud data. Based on the SalsaNext neural network, the SalsaNext++ neural network model is improved and proposed, and the SalsaNext++ neural network model (hereinafter referred to as the SalsaNext++ network) is improved and proposed. A new global context module is introduced in the SalsaNext++ network, named the diamond context module (DCB);

[0015] Step 1-2: Use the spherical projection method to map the point cloud data onto a two-dimensional spherical surface, and input the two-dimensional point cloud data into DCB to output context information of different scales;

[0016] Step 1-3: In the SalsaNext++ network coding stage, the features of the two-dimensional point cloud data are extracted through the dilated convolutional layer, gradually reducing the spatial dimension of the data, and capturing richer target features by increasing the number of channels;

[0017] Step 1-4: After the dilated convolutional layer of the SalsaNext++ network, an average pooling layer is used to further reduce the spatial resolution of the features while increasing the level of abstraction of the features to provide a more compact feature representation for the deep network;

[0018] Step 1-5: In the SalsaNext++ network decoding stage, the pixel shuffle layer is used to upsample the feature map output in the encoding stage to restore the spatial dimension of the data and provide the necessary spatial information for accurate semantic segmentation;

[0019] Step 1-6: The feature map of the encoding stage is fused with the upsampled features of the decoding stage through residual connection to enhance the network's ability to capture the local structure of the point cloud and improve the accuracy of the segmentation details;

[0020] Step 1-7: After feature fusion, use multi-scale convolutional layers to integrate and refine features of different scales to enhance the network's ability to understand the semantics of point clouds;

[0021] Steps 1-8: In the final stage of the SalsaNext++ network, the number of channels is adjusted again to match the number of semantic label types. The fused features are converted into the semantic category probability distribution of each point through the activation function layer to complete the semantic segmentation task of the point cloud.

[0022] Preferably, step 3 includes the following sub-steps:

[0023] Step 3-1: Set up a workspace in the ROS environment and create a new ROS package to encapsulate all related code and dependencies;

[0024] Step 3-2: Create and initialize a semantic segmentation node in the ROS environment. This node will be responsible for receiving the lidar point cloud data and performing semantic segmentation using the SalsaNext++ model.

[0025] Step 3-3: Load the trained SalsaNext++ model weight file into the semantic segmentation node and configure the semantic segmentation node parameters, including the topic name of the input point cloud data, the topic name of the output point cloud data with semantic labels, and the processing frequency;

[0026] Step 3-4: The semantic segmentation node subscribes to the topic of raw point cloud data published by the LiDAR node, and uses the SalsaNext++ model to perform semantic segmentation reasoning on the preprocessed point cloud data to obtain the semantic probability distribution of each point, and assigns a semantic label to each point accordingly;

[0027] Step 3-5: Publish the point cloud data with semantic labels to the specified ROS topic for use by the dynamic point cloud culling node.

[0028] Preferably, step 4 includes the following sub-steps:

[0029] Step 4-1: Create and initialize a dynamic point cloud culling node in the ROS environment, which is responsible for processing the point cloud data after semantic segmentation;

[0030] Step 4-2: Configure the parameters of the dynamic point cloud culling node, including the topic name of the subscribed point cloud data with semantic tags, the topic name of the published static point cloud data, the semantic tag threshold of the dynamic point cloud, and the processing frequency;

[0031] Step 4-3: Subscribe to the topic of point cloud data with semantic labels processed and published by the SalsaNext++ network through the dynamic point cloud culling node;

[0032] Step 4-4: For each point cloud data, check its semantic label. If the semantic label indicates that the point belongs to a dynamic object (such as a pedestrian or a moving vehicle), mark the point as a dynamic point cloud and remove it.

[0033] Step 4-5: After the dynamic point cloud removal node completes the removal of the dynamic point cloud, it publishes the processed static point cloud data to the specified topic for use by the laser radar synchronous positioning and mapping front-end module.

[0034] Preferably, step 5 includes the following sub-steps:

[0035] Step 5-1: Register the processed static point cloud to the laser radar coordinate system {L} to obtain the point cloud P obtained by the laser radar scan of the kth frame k ;

[0036] Step 5-2: Define i as the lidar point cloud P k A point in P, i∈P k , let S be the set of continuous points of i returned by the laser scanner in the same scan. Point cloud feature extraction is performed through different curvatures. In each scanning line of the laser, five points before and after a point are used to calculate the distance c from the average value to the point to characterize the curvature of the point, and plane points and edge points in the static point cloud are extracted. The specific formula is as follows:

[0037]

[0038] in, is the point cloud P of the kth scan in the L (radar) coordinate system k The i-th point in for Around the points.

[0039] Step 5-3: Perform feature matching, match the edge points and plane points of the current frame with the corresponding feature points in the previous frame to determine the correspondence between them. The matching process takes minimizing the distance from the edge point pair to the assumed straight line and the distance from the plane point to the assumed plane as the optimization goal;

[0040] Step 5-4: Use the Levenberg-Marquardt (LM) iterative solution method to estimate the position of the lidar, obtain the inter-frame position estimate of the lidar, and correct the motion distortion of the point cloud based on it to obtain the dedistorted static point cloud;

[0041] Step 5-5: Send the calculated inter-frame pose transformation of the LiDAR and the dedistorted static point cloud to the back-end optimization and map construction module for pose fusion estimation of the LiDAR's global pose in the world coordinate system and map update.

[0042] Preferably, step 6 comprises the following sub-steps:

[0043] Step 6-1: Initialize the global map, using the inter-frame pose transformation provided by the front-end odometer module and the corresponding dedistorted static point cloud data as the initial state;

[0044] Step 6-2: Apply the LM method to optimize the preliminary pose estimate provided by the front end. During the optimization process, the relative pose transformation between multiple consecutive frames of lidar data frames during the mapping process is considered, and it is combined with the inter-frame pose transformation provided by the front-end odometer to obtain the global pose estimate of the lidar relative to the map.

[0045] Step 6-3: Update the global map using the optimized global pose estimation, and register the static point cloud data after continuous multi-frame dedistortion provided by the front-end odometer to the map coordinate system to achieve map update and ensure the geometric consistency and accuracy of the map;

[0046] Step 6-4: In each iteration, evaluate the optimization results and decide whether the termination condition is met. If so, end the optimization process; if not, proceed to the next round of optimization and finally generate a high-precision static map.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] (1) High-precision semantic segmentation: The SalsaNext++ semantic segmentation network proposed in this paper can provide higher resolution and more accurate semantic segmentation by introducing an improved global context module (DCB), enhance the ability to express features of different scales in point cloud data, and improve the accuracy of segmentation detail information.

[0049] (2) Effective elimination of dynamic point clouds: The present invention adopts a semantic segmentation method based on the SalsaNext++ neural network, which can accurately semantically classify the lidar point cloud, effectively identify and eliminate dynamic objects such as moving vehicles and pedestrians, reduce the negative impact of the dynamic environment on the performance of the SLAM system, and improve the accuracy of point cloud registration and the quality of map construction.

[0050] (3) The dynamic point cloud in the laser point cloud data is effectively eliminated, thereby improving the accuracy of point cloud registration and point cloud map construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the process of the dynamic point cloud removal method in the present invention;

[0052] Figure 2 Schematic diagram of the semantic segmentation model architecture selected in the present invention;

[0053] Figure 3 This is a schematic diagram of the improved context module architecture of the present invention;

[0054] Figure 4 Schematic diagram of the LOAM algorithm structure for simultaneous positioning and mapping of lidar. DETAILED DESCRIPTION

[0055] 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 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 should fall within the scope of protection of the present invention.

[0056] The dynamic point cloud elimination method for laser radar synchronous positioning and mapping in this embodiment has the following process: Figure 1 As shown, including:

[0057] Step 1: Point cloud semantic segmentation network design: Considering hardware resource utilization, computing efficiency and training complexity, the SalsaNext neural network is selected as the basic model for the semantic segmentation of lidar point cloud data. Based on the SalsaNext neural network, the lidar point cloud data semantic segmentation network SalsaNext++ is designed; specifically:

[0058] Step 1-1: The SalsaNext network proposed in 2020 still ranks at the top of the list of semantic segmentation of point cloud in the Semantic-KITTI dataset. It does not require high hardware storage and computing power, has relatively low training complexity, and takes relatively little time to train. Therefore, the SalsaNext neural network is selected as the basic model for semantic segmentation of lidar point cloud data. Based on the SalsaNext neural network, the SalsaNext++ neural network model (hereinafter referred to as the SalsaNext++ network) is improved and proposed. The spherical projection method is used to map the point cloud data onto a two-dimensional sphere. At this time, the two-dimensional point cloud image is [w×h×5], where w and h represent the width and height of the image, such as Figure 2 It can be set to [2048×64×5];

[0059] Step 1-2: If Figure 3 As shown in the figure, a new global context module is introduced in the SalsaNext++ network, named Diamond Context Block (DCB). The two-dimensional point cloud data is input into DCB to output context information of different scales, with the number of channels from 5 to 32.

[0060] Step 1-3: In the SalsaNext++ network coding stage, the features of the two-dimensional point cloud data are extracted through the dilated convolutional layer, the spatial dimension of the data is gradually reduced, and more abundant target features are captured by increasing the number of channels. The number of channels changes from 32→64→128→256;

[0061] Step 1-4: After the SalsaNext++ network dilated convolutional layer, an average pooling layer is used to further reduce the spatial resolution of the features while increasing the abstraction level of the features to provide a more compact feature representation for the deep network. The point cloud size changes from 2048×64→512×32→256×16→128×8→64×4.

[0062] Step 1-5: In the SalsaNext++ network decoding stage, the pixel shuffle layer is used to upsample the feature map output in the encoding stage to restore the spatial dimension of the data and provide the necessary spatial information for accurate semantic segmentation. The point cloud size changes from 64×4→128×8→256×16→512×32→1024×64;

[0063] Step 1-6: The feature map of the encoding stage is fused with the upsampled features of the decoding stage through residual connection to enhance the network's ability to capture the local structure of the point cloud and improve the accuracy of the segmentation details;

[0064] Step 1-7: After feature fusion, use multi-scale convolutional layers to integrate and refine features of different scales to enhance the network's ability to understand the semantics of point clouds. The number of channels changes from 256 to 128 to 64 to 32.

[0065] Step 1-8: In the final stage of the SalsaNext++ network, in order to match the number of semantic label types, the number of channels is adjusted again, and the output is [2048×64×20]. Through the activation function layer, the fused features are converted into the semantic category probability distribution of each point to complete the semantic segmentation task of the point cloud.

[0066] Step 2: Point cloud semantic segmentation network training: Use the point cloud dataset Semantic-KITTI to train the semantic segmentation network, extract features and classify the input point cloud data, and assign the point cloud to predefined semantic categories such as vehicles, pedestrians, and buildings.

[0067] Step 3: Point cloud semantic segmentation: Integrate the trained SalsaNext++ semantic segmentation network into the LiDAR simultaneous positioning and mapping front-end system to process LiDAR point cloud data in real time, and identify dynamic point clouds (such as moving vehicles, pedestrians, etc.) and static point clouds (such as roads, buildings, etc.) through semantic segmentation. Specifically:

[0068] Step 3-1: Set up a workspace in the ROS environment and create a new ROS package to encapsulate all related code and dependencies;

[0069] Step 3-2: Create and initialize a semantic segmentation node in the ROS environment, named semantic_segmentation_node. This node will be responsible for receiving lidar point cloud data and performing semantic segmentation using the SalsaNext++ model.

[0070] Step 3-3: Load the trained SalsaNext++ model weight file into the semantic segmentation node, configure the semantic segmentation node parameters, set the topic name of the input point cloud data to Lidar_points, the topic name of the output point cloud data with semantic labels to Points_with_labels, and the processing frequency to 10Hz;

[0071] Step 3-4: The semantic segmentation node subscribes to the topic Lidar_points of the raw point cloud data published by the LiDAR node, and uses the SalsaNext++ model to perform semantic segmentation reasoning on the preprocessed point cloud data to obtain the semantic probability distribution of each point, and assigns a semantic label to each point accordingly;

[0072] Step 3-5: Publish the point cloud data with semantic labels to topic P o ints_with_labels, for use by the dynamic point cloud culling node.

[0073] Step 4: Dynamic point cloud removal: For the identified dynamic point clouds, according to the semantic labels added when assigned to the predefined semantic categories, they are removed through the point cloud removal algorithm and program, and only the static point clouds are retained for subsequent vehicle pose estimation and map construction. Specifically:

[0074] Step 4-1: Create and initialize a dynamic point cloud filter node in the ROS environment, named dynamicpoint_cloud_filter_node, which is responsible for processing the point cloud data after semantic segmentation;

[0075] Step 4-2: Configure the parameters of the dynamic point cloud rejection node. The topic name of the subscribed point cloud data with semantic labels is Points_with_1abels, the topic name of the published static point cloud data is Static_Lidar_points, the semantic label threshold of the dynamic point cloud is 252, and the processing frequency is 10Hz;

[0076] Step 4-3: Subscribe to the topic Points_with_labels of point cloud data with semantic labels processed and published by the SalsaNext++ network through the dynamic point cloud culling node;

[0077] Step 4-4: For each point cloud data, check its semantic label. If the data represented by the semantic label is greater than or equal to the threshold, indicating that the point belongs to a dynamic object (such as a pedestrian or a moving vehicle), then mark the point as a dynamic point cloud and remove it;

[0078] Step 4-5: After the dynamic point cloud removal node completes the removal of the dynamic point cloud, it publishes the processed static point cloud data to the topic Static_Lidar_points for use by the LiDAR synchronous positioning and mapping front-end module.

[0079] Step 5: Front-end odometer: In the front-end odometer module of the laser radar synchronous positioning and mapping system, the static point cloud data after removing the dynamic point cloud is used to quickly estimate the pose transformation between adjacent laser radar data frames through feature-based point cloud registration. The laser radar motion is estimated for point cloud dedistortion, providing inter-frame pose transformation and dedistorted static point cloud for laser radar synchronous positioning and mapping back-end optimization. Specifically:

[0080] Step 5-1: Register the processed static point cloud to the laser radar coordinate system {L} to obtain the point cloud P obtained by the laser radar scan of the kth frame k ;

[0081] Step 5-2: Define i as the lidar point cloud P k A point in P, i∈P k , let S be the set of continuous points of i returned by the laser scanner in the same scan. Point cloud feature extraction is performed through different curvatures. In each scanning line of the laser, five points before and after a point are used to calculate the distance c from the average value to the point to characterize the curvature of the point, and plane points and edge points in the static point cloud are extracted. The specific formula is as follows:

[0082]

[0083] in, is the point cloud P of the kth scan in the L (radar) coordinate system k The i-th point in for Around the points.

[0084] Step 5-3: Perform feature matching, match the edge points and plane points of the current frame with the corresponding feature points in the previous frame to determine the correspondence between them. The matching process takes minimizing the distance from the edge point pair to the assumed straight line and the distance from the plane point to the assumed plane as the optimization goal; Step 5-4: Use the Levenberg-Marquardt (LM) iterative solution method to estimate the position and posture of the lidar, obtain the inter-frame position and posture estimate of the lidar, and correct the motion distortion of the point cloud based on it to obtain the dedistorted static point cloud;

[0085] Step 5-5: Send the calculated inter-frame pose transformation of the LiDAR and the dedistorted static point cloud to the back-end optimization and map construction module for pose fusion estimation of the LiDAR's global pose in the world coordinate system and map update.

[0086] Step 6: Backend optimization and map construction: The backend optimization and map construction module receives the point cloud and inter-frame pose transformation after the front-end odometer has been distorted, performs global registration on the multi-frame point cloud, fuses the point cloud data into the world map for map update, and fuses the obtained global pose transformation with the inter-frame pose transformation to estimate the motion state of the lidar relative to the map. Specifically:

[0087] Step 6-1: Initialize the global map, using the inter-frame pose transformation provided by the front-end odometer module and the corresponding dedistorted static point cloud data as the initial state;

[0088] Step 6-2: Apply the LM method to optimize the preliminary pose estimate provided by the front end. During the optimization process, the relative pose transformation between 10 consecutive frames of lidar data frames during the mapping process is considered, and it is combined with the inter-frame pose transformation provided by the front-end odometer to obtain the global pose estimate of the lidar relative to the map.

[0089] Step 6-3: Update the global map using the optimized global pose estimation, and register the 10 consecutive frames of dedistorted static point cloud data provided by the front-end odometer to the map coordinate system to achieve map update and ensure the geometric consistency and accuracy of the map;

[0090] Step 6-4: In each iteration, evaluate the optimization results and decide whether the termination condition is met. If so, end the optimization process; if not, proceed to the next round of optimization and finally generate a high-precision static map.

[0091] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A dynamic point cloud elimination method for laser radar synchronous positioning and mapping, characterized in that: The following steps are involved: Step 1: Point cloud semantic segmentation network design: Taking into account hardware resource utilization, computing efficiency, and training complexity, the SalsaNext neural network is selected as the basic model for the semantic segmentation of lidar point cloud data. Based on the SalsaNext neural network, the lidar point cloud data semantic segmentation network SalsaNext++ is designed; Step 2: Point cloud semantic segmentation network training: Use the point cloud dataset Semantic-KITTI to train the semantic segmentation network, extract features and classify the input point cloud data, and assign the point cloud to predefined semantic categories; Step 3: Point cloud semantic segmentation: Integrate the trained SalsaNext++ semantic segmentation network into the LiDAR simultaneous positioning and mapping front-end system to process LiDAR point cloud data in real time and identify dynamic point clouds and static point clouds through semantic segmentation; Step 4: Dynamic point cloud removal: For the identified dynamic point clouds, according to the semantic labels added when they are assigned to the predefined semantic categories, they are removed through the point cloud removal algorithm and program, and only the static point clouds are retained for subsequent pose estimation and map construction; Step 5: Front-end odometer: In the front-end odometer module of the LiDAR simultaneous positioning and mapping system, the static point cloud data after the dynamic point cloud is removed is used to quickly estimate the pose transformation between adjacent LiDAR data frames through a feature-based point cloud registration method, and the motion of the LiDAR is estimated for point cloud dedistortion, providing inter-frame pose transformation and dedistorted static point cloud for LiDAR simultaneous positioning and mapping back-end optimization; Step 6: Backend optimization and map construction: The backend optimization and map construction module receives the dedistorted point cloud and inter-frame pose transformation of the front-end odometer, performs global registration on multiple frame point clouds, fuses the point cloud data into the world map for map update, and fuses the obtained global pose transformation with the inter-frame pose transformation to estimate the motion state of the lidar relative to the map.

2. The dynamic point cloud elimination method for laser radar synchronous positioning and mapping according to claim 1 is characterized in that: The step 1 includes the following sub-steps: Step 1-1: Based on the SalsaNext neural network, the SalsaNext++ neural network model is improved and proposed, referred to as the SalsaNext++ network. A new global context module is introduced into the SalsaNext++ network, named the diamond context module DCB; Step 1-2: Use the spherical projection method to map the point cloud data onto a two-dimensional spherical surface, and input the two-dimensional point cloud data into DCB to output context information of different scales; Step 1-3: In the SalsaNext++ network coding stage, the features of the two-dimensional point cloud data are extracted through the dilated convolutional layer, gradually reducing the spatial dimension of the data, and capturing richer target features by increasing the number of channels; Step 1-4: After the dilated convolutional layer of the SalsaNext++ network, an average pooling layer is used to further reduce the spatial resolution of the features while increasing the level of abstraction of the features to provide a more compact feature representation for the deep network; Step 1-5: In the SalsaNext++ network decoding stage, the pixel shuffle layer is used to upsample the feature map output in the encoding stage to restore the spatial dimension of the data and provide the necessary spatial information for accurate semantic segmentation; Step 1-6: The feature map of the encoding stage is fused with the upsampled features of the decoding stage through residual connection to enhance the network's ability to capture the local structure of the point cloud and improve the accuracy of the segmentation details; Step 1-7: After feature fusion, use multi-scale convolutional layers to integrate and refine features of different scales to enhance the network's ability to understand the semantics of point clouds; Steps 1-8: In the final stage of the SalsaNext++ network, the number of channels is adjusted again to match the number of semantic label types. The fused features are converted into the semantic category probability distribution of each point through the activation function layer to complete the semantic segmentation task of the point cloud.

3. The dynamic point cloud elimination method for laser radar synchronous positioning and mapping according to claim 2 is characterized in that: The step 3 includes the following sub-steps: Step 3-1: Set up a workspace in the ROS environment and create a new ROS package to encapsulate all related code and dependencies; Step 3-2: Create and initialize a semantic segmentation node in the ROS environment. This node will be responsible for receiving the lidar point cloud data and performing semantic segmentation using the SalsaNext++ model. Step 3-3: Load the trained SalsaNext++ model weight file into the semantic segmentation node and configure the semantic segmentation node parameters, including the topic name of the input point cloud data, the topic name of the output point cloud data with semantic labels, and the processing frequency; Step 3-4: The semantic segmentation node subscribes to the topic of raw point cloud data published by the LiDAR node, and uses the SalsaNext++ model to perform semantic segmentation reasoning on the preprocessed point cloud data to obtain the semantic probability distribution of each point, and assigns a semantic label to each point accordingly; Step 3-5: Publish the point cloud data with semantic labels to the specified ROS topic for use by the dynamic point cloud culling node.

4. The dynamic point cloud elimination method for laser radar synchronous positioning and mapping according to claim 3 is characterized in that: The step 4 includes the following sub-steps: Step 4-1: Create and initialize a dynamic point cloud culling node in the ROS environment, which is responsible for processing the point cloud data after semantic segmentation; Step 4-2: Configure the parameters of the dynamic point cloud culling node, including the topic name of the subscribed point cloud data with semantic tags, the topic name of the published static point cloud data, the semantic tag threshold of the dynamic point cloud, and the processing frequency; Step 4-3: Subscribe to the topic of point cloud data with semantic labels processed and published by the SalsaNext++ network through the dynamic point cloud culling node; Step 4-4: For each point cloud data, check its semantic label. If the semantic label indicates that the point belongs to a dynamic object, mark the point as a dynamic point cloud and remove it. Step 4-5: After the dynamic point cloud removal node completes the removal of the dynamic point cloud, it publishes the processed static point cloud data to the specified topic for use by the laser radar synchronous positioning and mapping front-end module.

5. The dynamic point cloud elimination method for laser radar synchronous positioning and mapping according to claim 1, characterized in that: The step 5 comprises the following sub-steps: Step 5-1: Register the processed static point cloud to the laser radar coordinate system {L} to obtain the point cloud P obtained by the laser radar scan of the kth frame k ; Step 5-2: Define i as the lidar point cloud P k A point in P, i∈P k , let S be the set of continuous points of i returned by the laser scanner in the same scan, and extract point cloud features by different curvatures. In each scanning line of the laser, five points before and after a point are used to calculate the distance c from the average value to the point to characterize the curvature of the point, and extract the plane points and edge points in the static point cloud. The specific formula is as follows: in, is the point cloud P of the kth scan in the L radar coordinate system k The i-th point in for The surrounding points; Step 5-3: Perform feature matching, match the edge points and plane points of the current frame with the corresponding feature points in the previous frame to determine the correspondence between them. The matching process takes minimizing the distance from the edge point pair to the assumed straight line and the distance from the plane point to the assumed plane as the optimization goal; Step 5-4: Use the Levenberg-Marquardt iterative solution method to estimate the position of the lidar, obtain the inter-frame position estimate of the lidar, and correct the motion distortion of the point cloud based on it to obtain the dedistorted static point cloud; Step 5-5: Send the calculated inter-frame pose transformation of the LiDAR and the dedistorted static point cloud to the back-end optimization and map construction module for pose fusion estimation of the LiDAR's global pose in the world coordinate system and map update.

6. The dynamic point cloud elimination method for laser radar synchronous positioning and mapping according to claim 1, characterized in that: The step 6 comprises the following sub-steps: Step 6-1: Initialize the global map, using the inter-frame pose transformation provided by the front-end odometer module and the corresponding dedistorted static point cloud data as the initial state; Step 6-2: Apply the LM method to optimize the preliminary pose estimate provided by the front end. During the optimization process, the relative pose transformation between multiple consecutive frames of lidar data frames during the mapping process is considered, and it is combined with the inter-frame pose transformation provided by the front-end odometer to obtain the global pose estimate of the lidar relative to the map. Step 6-3: Update the global map using the optimized global pose estimation, and register the static point cloud data after continuous multi-frame dedistortion provided by the front-end odometer to the map coordinate system to achieve map update and ensure the geometric consistency and accuracy of the map; Step 6-4: In each iteration, evaluate the optimization results and decide whether the termination condition is met. If so, end the optimization process. If it is not satisfied, continue to the next round of optimization and finally generate a high-precision static map.

Images