LiDAR-Based Mobile Robot Outdoor Place Recognition And Map Maintenance

Place recognition tasks enable mobile robots to recognize scenes previously visited and play an important role in autonomous localization tasks for large-scale outdoor environments.However,in the outdoor environment,there are complex environmental conditions such as lighting,seasons,and even artificial scene reconstruction.These environmental changes will adversely affect the mobile robot's environmental perception,which will affect the accuracy and stability of the place recognition task.Severe environmental changes not only cause the failure of place recognition task,but also cause the historical data of the corresponding scenes in the priori map to become outdated and invalid.Therefore,mobile robots working in outdoor environments for a long time must not only improve the ability of place recognition to adapt to environmental changes,it should also have the ability to update and maintain the priori map according to environmental changes to ensure the timeliness of the priori map and the stability of autonomous localization.In order to improve the adaptability of the mobile robot place recognition and mapping task to the complex changes in the outdoor environment,this thesis conducted the following innovative research:(1)Nowadays although most of the state-of-the-art place recognition methods are vision-based,the visual sensors lack adaptability to light changes and poor light conditions,so this thesis proposes a solution based on LiDAR.Although some research results of lidar-based place recognition have been achieved at this stage,how to efficiently extract scene feature descriptions from 3D LiDAR point clouds has not been effectively solved.In response to this problem,a panoramic bearing angle image model is proposed to realize the image representation of the 3D LiDAR point cloud.Extracting the feature description of the scene from the image can effectively reduce the complexity of the algorithm.Based on the panoramic bearing angle image model,this work uses the bag of words algorithm of visual features to achieve autonomous place recognition.In addition,the geometric constraints of feature matching pairs can effectively improve the accuracy of place recognition results.Experimental verification based on multiple datasets verifies the effectiveness of the method.(2)In order to improve the adaptability to seasonal changes in outdoor environment,a solution based on scene context is proposed.Through the analysis of the cross-season LiDAR point cloud data,it is found that the information such as the shapes and layout relationship of objects in the scene maintains a certain stability during the season shfts process.In this work,to avoid the computational complexity of cluster analysis of point clouds in 3D space,a novel ring-view representation model of 3D scene is proposed.The global texture features based on the scene ring view can efficiently extract and describe the shapes and layout relationship of objects in the scene.In addition,to improve the accuracy of place recognition results,matching between scene sequences is used instead of matching between single scenes.Experimental verification based on three cross-season datasets verifies the effectiveness of the proposed method.(3)To solve the problem of timeliness of a priori map of an outdoor mobile robot,a long-term maintenance method of topological maps based on multiple scene sampling is proposed.During repeated visits,mobile robots constantly relocate the currently constructed topological nodes in the prior topological map,and use the relocation results to determine whether the current scene has changed,and add the changed scene as a new topological node to priori map.The relocation method based on cross-seasonal autonomous place recognition is robust to environmental changes and can effectively reduce the number of new topological nodes.At the same time,in order to ensure the accuracy of the relocation result,a method for checking the relocation result based on the consistency constraint of topological structure is proposed.Experimental verification based on three datasets with a construction period of more than one year verifies the effectiveness of the proposed method.Experimental results show that updating the map can effectively improve the stability of mobile robot relocation.