Research On Collaborative Area Exploration Technology Based On Sparse Point Cloud

In recent years,with the extensive application and development of agents in military and civil fields,more and more tasks such as reconnaissance,rescue and combat can be well accomplished by agents.In an unknown environment,it is very important to accurately estimate the position and attitude of an agent and obtain information about the surrounding environment to navigate it.At the same time,when a single agent is exploring a large unknown environment,its efficiency is often low due to limited information sources.Therefore,how to use multiple agents to efficiently complete the mapping and exploration is undoubtedly the hot spot of current research.The main research content of this topic is to realize efficient collaborative search mapping for unknown regions based on visual SLAM and regional exploration technology.In this project,the depth camera is used as the visual sensor,sparse point cloud is used as the input,combined with the path planning algorithm,and unmanned vehicle is used to carry out collaborative exploration and mapping of the designated area.In the process of exploration,the advantages of multiple agents are brought into play to make the overall intelligent unmanned system both efficient and accurate.Relevant research methods are as follows:Firstly,a collaborative vision SLAM algorithm based on depth camera is designed.This algorithm extracts ORB feature points and matches FLANN feature points on each robot,and adopts the fast average method combined with RANSAC algorithm to eliminate the wrong matches.Then,the singular value decomposition method is used to solve the ICP problem to estimate the pose and obtain the respective sparse point cloud map.The word bag model is used to match the overlapping areas of each feature point cloud,so that the information collected by multiple robots can be combined to generate a complete sparse point cloud map.Secondly,the transformation process from sparse point cloud map to two-dimensional raster map was designed.First,sparse point cloud was sampled by EAR algorithm,then a three-dimensional octree map was formed by finite space equalization,and finally a two-dimensional raster map was generated by highly filtering and projection for mobile robot navigation.Thirdly,the multi-agent collaborative region exploration algorithm is designed.The random tree is rapidly expanded to search for the target points to be selected on the robot's path,the mean offset clustering is used to preprocess the target points,and then the target points are generated according to the profit function judgment and the multi-machine task assignment algorithm.Finally,A~* algorithm and DWA algorithm are used to realize point-to-point path planning.While the robot moves,it continues to collect information to build maps,and finally realizes the exploration task of unknown areas.Finally,the overall design of area exploration was implemented under the Ubuntu system,and the simulation scene was built in Gazebo to test the accuracy,stability and efficiency of the system.Meanwhile,based on the characteristics of sparse point cloud,the system is improved to improve the performance of the system.