Research On SLAM Algorithm Based On Depth Camera

Now,intelligent mobile robot is the main research direction of many scholars,and the SLAM technology is the key for mobile robot to achieve independent movement.VSLAM(Visual SLAM)system is the technology that can localization and mapping simultaneously with camera as the main sensor.VSLMA system is the main research object in the field of SLAM technology because of its high price performance ratio and abundant information.The SLAM system with depth camera among VSLMA system can obtain color image and depth image correspond one by one at the same time,which brings great convenience to the research of SLAM system.However,there are many shortcomings in the existing VSLAM system.For example,the estimation of camera pose is not accurate enough,and the overall optimization of back-end takes too long,which limits the speed of the robot.Based on the existing SLAM system model,depth camera is used to improve the design of front-end visual odometry and overall optimization of back-end in this paper.For the part of front-end,this paper focuses on improving the accuracy of estimation of camera pose without affecting the real-time performance.In this paper,the feature extraction algorithm----ORB algorithm is used to extract features of color image and calculate descriptors.At the same time,the local map is constructed according to the threshold value of keyframes.Then,the feature match between the local map and the current color image is carried out ac cording to the descriptors.Finally,Pn P algorithm is used to calculate the camera pose according to the correct results of feature match.In order to ensure the robustness of the localization process,the triangulation used to update the depth of map points in the local map,after fusing the value from the triangulation and the value from the depth image.The more accurate the depth information,the better the effect of feature match,so as to improve the accuracy of estimation of camera pose.For the part of back-end optimization,the BA optimization algorithm is improved in this paper,and become a back-end optimization algorithm with better real-time performance.It improves the shortcomings of BA optimization algorithm,such as the gradual expansion of scene,the sharp increase of calculation,and the impact on real-time performance.In this method,the importance of camera points and waypoints is sorted by HITS algorithm firstly.And then the nodes with poor correlation are deleted,so as to reduce the number of nodes to be optimized in the back-end and improve the real-time performance.In the end of this paper,the fr1 data set in the TUM database were used to test and verify the change of accuracy of visual odometry and the effect of the back-end overall optimization.