Research On Visual SLAM System In Dynamic Environment

Environmental perception and navigation is one of the basic problems for robotics Simultaneous localization and mapping(SLAM)is an important method for robots to perceive the environment and obtain their own pose.With the advantages of low cost and rich information,visual sensor has been widely used in SLAM and has gradually become a research hotspot.At present,most of the research in visual SLAM can only work in the static scene.If there is a moving object in the image,it will affect the accuracy and robustness of SLAM system.Therefore,this paper mainly studies visual SLAM in dynamic environment.In the dynamic environment,our method can estimate the pose of the camera and the moving objects in real time.The main research work and contributions are summarized as follows:First,in order to solve the problem of pose estimation in dynamic environment,this paper proposes a movement segmentation algorithm based on motion consistency.Firstly,the current image and the previous image are feature matched and compute the velocity of each feature point.Next,triangulate the feature points to create connections between them.Then,according to the consistency of the feature points' velocities,the unstable connections are eliminated and the feature points are divided into dynamic groups and static groups.Finally,only static feature points are used to estimate camera pose.Second,in order to solve the problem of back-end optimization and loop closure in dynamic environment,this paper proposes a method of dynamic point cloud segmentation based on local map consistency and a method of loop detection based on occlusion compensation of moving objects.The method of dynamic point cloud segmentation based on local map consistency is to eliminate the dynamic point cloud according to the observation of key frames in a period of time.This method effectively reduces the influence of dynamic point cloud and improves the accuracy and robustness of back-end optimization.The method of loop detection based on occlusion compensation of moving objects is to compensate the occluded area of moving objects in the current key frame through the information of adjacent key frames.This method improves the accuracy and robustness of loop closure in the dynamic environment.Experiments show that,the visual SLAM system proposed in this paper has good performance in dynamic environment,compared with ORBSLAM,DVO and SPWSLAM.Thirdly,in order to solve the problem of moving objects' pose estimation,this paper proposes a method of tracking and moving objects' pose estimation based on feature points.In this paper,movement segmentation,camera's pose estimation and moving objects' pose estimation are integrated into a unified visual SLAM framework.Therefore,the system can estimate the pose of moving objects in real time without prior information.