Research On Robot Grid Map Construction Based On Vision And Laser Sensor Fusion

With the progress of social science and the improvement of production and labor level,the demand for robots is also advancing.With the continuous improvement of robot technology,more and more mobile robots have entered various fields of society,helping people to complete various tasks to meet people’s needs.Mobile robot can accurately and real-time locate its position in the unknown complex environment and build an accurate and complete environment map,which is the key technology for mobile robot to complete the multi-functional deployment task.However,mobile robots only rely on a single sensor,which has great shortcomings and limitations in realizing simultaneous positioning and map construction(SLAM).Aiming at the problem that the scanning range of lidar sensor is limited to the environmental information of two-dimensional plane,and there is inaccurate map information due to the obstacle information that cannot be scanned during map construction,this paper realizes the map construction in SLAM Based on robot operation system(ROS)and adopts the method of multi-sensor information fusion of lidar and rgbd camera Kinect v1.1.Lidar uses RBPF(Rao-Blackwillised Particle Filter)SLAM based algorithm and Hector SLAM algorithm based on scan matching to build maps.The problem of location accuracy drop caused by RBPF algorithm and the lack of diversity caused by multiple resamples of particles are improved.The improved resampling algorithm introduces a low weight particle set based on the premise of multiple copies of high weight particles,which improves the problem that the map updates only rely on a few particles and the diversity of particles is gradually lost in the multiple resampling process.By comparing HectorSLAM,original particle filter SLAM and improved resampling RBPFSLAM in the simulation environment,it is verified that the improved algorithm improves the accuracy of lidar registration and composition algorithm.2.The vision sensor RGBD camera Kinect carries out the construction of three-dimensional point cloud.It studies and improves the problems of inaccurate feature extraction,slow feature matching speed and the inability of three-dimensional point cloud to accurately construct the surrounding environment information in-the traditional point cloud construction.It improves the feature point extraction,uses RANSAC algorithm for feature matching,adds loop detection and point cloud splicing to improve the accuracy of three-dimensional point cloud construction,and improves the accuracy of three-dimensional point cloud construction through TUM data set fr1_xyz and real scene experiments are used to verify,and the grid map of visual slam is constructed based on ORBSLAM2.3.In reality,the scene construction relies on ROS to control Turtlebot mobile robot for map construction experiment.Use a single lidar to build a two-dimensional grid occupation map of the real scene,then use a single depth camera Kinect to build a three-dimensional point cloud model through the visual SLAM,and use the three-dimensional information obtained by the Kinect camera to build a two-dimensional grid occupation map of the visual SLAM based on ORBSLAM2.Based on Bayesian estimation fusion theory and occupancy grid map construction algorithm,multi-sensor information fusion map construction is realized.According to the construction result of real scene map,the environment map constructed by multi-sensor information fusion is more accurate and complete than that constructed by single sensor.