Lidar-based Indoor Obstacle Detection And Status Estimation

The environment perception technology of indoor scene is one of the most concerned research contents in the field of service robot research in recent years.The research of indoor service robots in China started relatively late,although it has been applied in some fields,but a lot of research is still theoretical derivation and experimental simulation.Lidar can be direct and accurate when obtaining the depth information of obstacles,and at the same time can ensure a certain scanning range,which has good results in the process of detecting indoor obstacle targets and estimating the state.In this paper,the obstacle detection and state estimation method using 3D lidar is studied for the indoor work scene of the robot with moving obstacles.Research is carried out from the aspects of point cloud data processing,obstacle point cloud clustering,and moving obstacle state estimation.Specific contents include:(1)Build a lidar indoor obstacle detection platform,based on the Ubuntu16.04 operating system,combined with ROS platform and PCL point cloud library,build a lidar obstacle detection software environment.The carrier trolley and lidar are fixed and installed to simulate the indoor work scene of the service robot in the laboratory.C++ language programming is used to realize the original data acquisition and subsequent processing process of Li DAR.(2)Study the processing algorithm of invalid point cloud in lidar data.The coordinate conversion relationship between lidar and carrier trolley was analyzed,and the original laser point cloud data under the coordinate system of the carrier trolley was obtained.Through the installation and calibration of the lidar,the rotation and translation parameter matrix is obtained,and the coordinate transformation between the two is realized.The preprocessing of the original point cloud data is realized by statistical filtering and voxel filtering,and the ground data is fitted and removed by the RANSAC algorithm.(3)Aiming at the problem of poor real-time performance of DBSCAN clustering algorithm in the process of obstacle point cloud clustering,an improved DNBSCAN clustering algorithm using KD-tree acceleration is studied.On the basis of the traditional DNBSCAN clustering algorithm,the selection interval of two important parameters of the algorithm is calculated,and the traditional DNBSCAN clustering algorithm is improved by taking advantage of the fast search data of KD-tree.Experiments show that the algorithm proposed in this paper can improve the real-time performance of clustering.(4)Aiming at the problem of dynamic obstacles in the indoor environment,the motion state estimation method of dynamic obstacles was studied.Set pedestrians walking randomly in the indoor environment to simulate dynamic obstacle movement scenarios,and realize data correlation of the same obstacle at different moments by establishing correlation gates and combining joint probability data correlation algorithms.The Kalman filter is used to estimate the position and velocity information of obstacles respectively,and the experimental results show that the state estimation of obstacles can be achieved within the error range.