Research On Methods Of Realtime Location And Mapping And Autonomous Exploration For UAV In Underground Mine

There are lots of cavity,stope and risky tunnel in underground mine of our country,with the mining and the influence of the pressure disturbance,these regions exist in roof caving,wall caving,and even the risk of collapse,which will seriously affect the normal production of mining enterprises and people’s life and property safety,it is urgent to develop technological means to proven its internal threedimensional shape.UAV is an ideal platform for underground mine detection in restricted environment due to its small size,good mobile performance,strong hovering ability and other advantages.In order to realize the mission of flight detection in the dangerous area of underground mine,it is crucial for the UAV to be able to fly autonomously to reduce the risk of operators and improve the detection efficiency.However,the underground mine has the characteristics of no GPS positioning signal,complex spatial topological structure and distribution of small obstacles,the UAV is faced with many problems such as difficult positioning and mapping,low efficiency of autonomous exploration,and high flight safety risk when it wants to realize autonomous flight in the underground mine.In order to effectively solve these problems of autonomous flight of UAV in underground mine,this paper studies the data pre-processing collected during UAV flight,real-time positioning and mapping of UAV in the absence of GPS signals,and autonomous exploration of UAV in the complex space structure.The main contents include:(1)Aiming at the problem of distortion of laser frame point cloud data collected by the external composite motion,a method for removing the distortion of multi-line LiDAR point cloud data under the condition of composite motion was proposed.Compared to the distortion removal method based on uniform motion,The method in this paper uses the inertial measurement unit to interpolate the laser frame point cloud data by spherical interpolation,and then compensates the external motion of the point cloud data to effectively correct the motion distortion of the point cloud data The experiment results show that the proposed method can remove not only rotational motion distortion but also the combined motion distortion of translation and rotation.(2)To solve these problems of pose estimation and map construction failure caused by lack of features of underground mines and sparse point cloud data,a realtime pose estimation and map construction method of UAV based on probability features and spatial features was proposed.Firstly,an adaptive grid is constructed along the main direction of the laser frame point cloud data,so that the point cloud inside each grid can reflect the local distribution.Secondly,a probabilistic featurebased matching method is used to achieve the inter-frame matching of the laser to ensure that the UAV pose can be stably estimated in the area of lack of underground mine features.Then,the estimated pose is used as the initial value of the local laser frame and the spatial features-based method is used to construct the environment map dynamically.The experiment results show that the R-LIO-PS algorithm proposed in this paper is superior to the current mainstream FASTER-LIO algorithm,FAST-LI02 algorithm and LIO-SAM-ODOM algorithm in terms of the mapping effect,point cloud accuracy and flight positioning accuracy in both underground mine stope and underground mine tunnel.(3)To solve these problems that the real-time estimated pose and the constructed map will drift over time or with the increase of the moving distance,a pose and map optimization method based on repeated scene constraints was proposed.Because of the large amount of laser frame data,the strategy of laser keyframe extraction,laser keyframe nearest neighbor iteration and similarity estimation is used to detect the repeated scene quickly.Based on the detected repeated scenes and the existing pose information,the global pose optimization model is constructed,and the global update of historical pose and map is realized.The experiment results show that the absolute error of measuring points,relative error between measuring points and positioning error increase by 93.1%,93.3%and 92.6%respectively compared with the LIO-SAM algorithm after optimizing the data of underground long-distance tunnel using the R-LIO-RS algorithm proposed in this paper.(4)To solve these problems of low efficiency and high flight risk caused by many branch channels,single head area and small obstacles in underground mines,an autonomous exploration method based on local generalized Voronoi diagram constraint was proposed.Compared with the traditional fast search random tree method,the constructed fast search random graph can select feasible candidate paths in underground mines with many branch channels.The strategy of local path planning and global path planning is adopted to quickly select a flight path from the candidate paths when the UAV is stuck in stagnation.The flight path is optimized based on the fact that the distance between local generalized Voronoi diagram boundary and any point in Voronoi diagram element is the same,so that the UAV can pass safely in the area of narrow underground mine space and small obstacles.The experiment results show that LVGPlanner algorithm proposed in this paper is superior to NBVPlanner algorithm in exploration efficiency and safety in both simulation scenarios and actual underground mine scenarios.