Research On Application Of Visual SLAM In Dynamic Environment

With the increase of the elderly population in my country,the demand for old-age care has increased significantly,so indoor service robots have received more and more attention.In the past few decades,whether it is in the field of computer vision or the robot community,visual simultaneous localization and mapping(visual SLAM)technology has aroused great research interest.And there are many algorithms for visual SLAM in dynamic environment.However,the practical application of the algorithms still faces the problem of insufficient robustness and real-time performance.In response to these problems,this article studies the practical application of the visual SLAM algorithm in a dynamic environment.The main contents of this article are as follows:(1)With the aim of eliminating dynamic objects in the environment,an RGB-D SLAM system is proposed.The ORB features in the image are selected to increase the speed of the system.The back-end optimization and loop detection algorithms are added to improve the robustness of the system and eliminate accumulated errors.The key frame technique is used to reduce the calculation cost.The Mask R-CNN algorithm is fused into the system to eliminate predefined dynamic objects.In order to remove dynamic objects that are not predefined,a dynamic object removal algorithm based on geometric constraints is proposed.(2)An algorithm for fusing visual SLAM with wheel speed encoder is proposed.Through the establishment of the odometry model of the two-wheeled robot,the connection between the wheel speed encoder and the robot pose is found,and the front end of the visual SLAM is fused to jointly optimize the robot's pose.Combined with the deep neural network Mask R-CNN algorithm,the predefined moving objects in the image are removed.Introduce K-means algorithm,combined with image depth,to eliminate unpredefined dynamic pixels.Using key frames,the system octree map is constructed.(3)Based on the Turtle Bot2 robot,combined with the Kinect 2 camera,a mobilerobot experiment platform was built,and the code of the system proposed in this paper was designed in conjunction with the ROS robot operating system.And tested the proposed SLAM system in the laboratory environment,and analyzed the performance of the system on the actual wheeled robot.