Research On Localization And Gapping Of Binocular Visual SLAM In Dynamic Environment

With the rapid development of artificial intelligence,visual SLAM(Simultaneous Localization and Mapping)technology combined with deep learning has become a hot research direction in recent years.Visual SLAM enables mobile platforms to construct maps of unknown surroundings in real time with only visual sensors,which is a key technology for mobile platforms to realize autonomous navigation.However,due to the complexity of the scene caused by moving objects and image noise in the environment,the positioning and mapping effect of the system are seriously affected.Therefore,in order to improve the accuracy of pose estimation for visual SLAM in dynamic environment and enhance the robustness of the system,this paper proposes a visual localization method that eliminates the influence of moving object surface features in dynamic scene,and studies the construction of dense point cloud map of the scene.The main research contents are as follows:Firstly,aiming at the interference of moving objects to camera pose estimation in dynamic environment,a method of feature elimination in dynamic region was proposed to obtain the image position of detected objects by combining semantic information,and the uniform distribution and accurate matching of feature points were realized by using the quadtree model and the matching,screening and elimination method.Secondly,in order to solve the problem that sparse maps are difficult to meet the navigation obstacle avoidance,a real-time dense point cloud map construction method based on binocular stereo matching was proposed.In addition,by using the neighborhood pixel value,the void in the disparity map was filled,which effectively enhanced the texture information of the point cloud map.Thirdly,through the evaluation of location and mapping effect on the public datasets,it is verified that the proposed algorithm can effectively improve the positioning accuracy of the system in the dynamic environment,and construct the environment dense point cloud map with clear texture.