Research On Visual SLAM Algorithm In Dynamic Scene

With the rapid development of artificial intelligence technology,the use of intelligent robots is becoming more and more extensive.SLAM(Simultaneous Localization and Mapping)technology,as an important method to assist intelligent robot to achieve autonomous positioning and navigation,has been paid attention by many researchers."Visual SLAM" refers to the SLAM technology that uses visual sensors to obtain the external environment information.At present,most visual SLAM algorithms work well in static environments,however,if there are dynamic objects in the scene,it will make the positioning accuracy of the system have a large deviation,affect the mapping effect,and even lead to system failure.Therefore,this paper studies the visual SLAM method in dynamic scenes,and improves the traditional visual SLAM framework,aiming to improve the positioning accuracy and mapping effect of visual SLAM in a dynamic environment.The main contents are as follows:(1)Combine the deep learning network and the motion compensation frame difference method to propose a dynamic target detection algorithm.This method can reduce the noise generated by the static background of the motion compensation frame difference method when the image is blurred,strong parallax,and dynamic objects occupy a large proportion of the image,and can more effectively detect dynamic objects.(2)The moving object detection method is applied to visual SLAM,and a visual SLAM system suitable for dynamic scene is proposed.This system improves the traditional visual SLAM process framework,by adding dynamic object detection module,effectively reduces the impact of dynamic objects in the environment on visual SLAM,and improves the robustness of SLAM system in dynamic environment.(3)The idea of moving object detection is applied to SLAM mapping,and a visual SLAM dense point cloud mapping method is proposed.Aiming at the problem that dynamic objects will produce noise in dense point cloud mapping,this method improves the point cloud mapping thread,and uses deep learning network to filter out dynamic objects,so as to reduce the impact of dynamic objects and improve the readability of the map.