Technical Implementation Of ROMO Map Reconstruction Based On RGBD

With the rapid development of robotics and AI technology,the demands on the precision of robot navigation and localization in application are increasing.Therefore,a much more accurate environmental map is needed.The current RGB-D map boasts of high accuracy,delicate structure and powerful interactive features,but with poor anti-interference performance.In order to deal with the problem of noise interference in dynamic environments,this thesis sets up the framework on the basis of RGB-D SLAM,applies Kinect as the main information acquisition equipment,and finally builds a global consistent point cloud map.Herein,the writer takes into comprehensive consideration the background code and pixel consistency to extract environmental background points.The main steps are summarized as follows:(1)Calibrated and processed the internal and external parameters of Kinect and then established a unified map coordinate system.On account that depth data collected by Kinect are easily influenced by environment,the deep preprocessing module needs to be designed to restore the data so as to improve the map's accuracy.(2)In the dynamic condition,a moving object will have more than one image during RGB-D SLAM map building,which seriously affects the map's quality.Therefore,this thesis has proposed building a vision-based static background point cloud map.Firstly,the author set up a code model after plenty of background points marking,segmenting all moving background points from those foreground ones in the keyframe.And then,the author fixed the right static background point through overall analysis on inter-frame gray scale variation and reprojection depth of potential static background points.Finally by means of Iterative Closest Point(ICP),the global consistent point cloud was gotten.(3)Built a ROS frame-based mobile platform,simulated robot motion and then set up dynamic laboratory environmental map by applying RGB-D SLAM and Kinect.The author has validated the removal of the moving object and static point cloud fusion capability and compared real trajectories with the simulation ones in an equal-scale laboratory model environment.The results of a series of experiments have showed that background map built in accordance with the method in this thesis is of higher applicability and positioning accuracy.