Research On Mptimization Method Of Simultaneous Location And Mapping Based On Vision

SLAM(Simultaneous Localization and Mapping)algorithm has been applied to intelligent robots,drones,AR/VR,driverless and other fields,but the current SLAM algorithms can still be optimized in terms of operating speed,camera trace accuracy,and robustness.In order to improve the accuracy of camera trajectories in SLAM,SLAM algorithm based on sparse direct method is proposed.The improved Shi-Tomasi feature point detection algorithm is used in the front-end to extract feature points.And monocular initialization is performed using an error-based initialization method.Based on the extracted feature points,the direct pose method is used to estimate the pose of the camera,and the constructed map is used to optimize the pose.In the back end,the optimization of the depth value,local and loop detection modules is used to reduce camera pose errors.According to the TUM standard dataset,through comparative analysis,the SLAM algorithm based on sparse direct method can effectively reduce the error.The average camera trajectory error is 4% of the map size,which is better than the ORB-SLAM2 algorithm.