Simultaneous Localization And Mapping For Robot Based On Point-line Feature Fusion

Simultaneous localization and mapping(SLAM)is one of the key technology in the field of mobile robot,which is the premise of autonomous mobile robot and navigation,and has a wide range of application prospects.At present,visual SLAM(v SLAM)is mostly based on the point features in real scene.However,the point features are very dependent on the environment,and there will be many challenges in the low-texture scene.There are lots of line features in the structured environment,which can make up for the shortcomings of point features.In this paper,a visual SLAM system based on point-line feature fusion is proposed by using depth camera,which can make up for the deficiency in low-texture scene and improve the robustness of SLAM system.The main contents of this paper include:(1)Firstly,we introduce the research significance and background,and analyzes the development history of SLAM and the research status of visual SLAM.Then,we introduce the related technology of v SLAM,including the state estimation,models of camera,and finally introduce the v SLAM framwork,including visual odometry,backend optimization,loop closing and mapping.(2)This paper proposes an RGB-D SLAM method based on point-line fusion,which aims to solve the problems of insufficient point features,motion jitter,low texture and so on in the method based on point features.Firstly,an adaptive line segment extraction method is proposed to solve the overlap or branch problem of the troditional line segments method;Secondly,a more rigorous screening mechanism is proposed in the line matching section;Instead of minimizing the reprojection error of points,we introduce the reprojection error based on points and lines to get more accurate tracking pose.In addition,we come up with a solution to handle the jitter frame,which greatly improves tracking success rate and availability of the system.We thoroughly evaluate our system on the TUM RGB-D benchmark and compare it with ORB-SLAM2.The experiments show that our system has better accuracy and robustness compared to the ORB-SLAM2.(3)Keyframe has a constraint on camera pose optimization.In this paper,a keyframe selection based on mutiple mechanisms is proposed to overcome the shotcomings of common keyframe selection methods.Firstly,combined with the matching of map points and line features,a keyframe filtering strategy based on convisibility graph is proposed to get the first level keyframe,which is used for local pose optimization and loop closure.Then based on the first level keyframe,a distance evaluation formula is defined to get the second level keyframe,which is used to realize the real-time construction of point cloud map in this chapter.Compared with sparse map,dense point cloud map can fully reflect the scene structure.At last,the experiments on the keyframe trajectory error,number and the effect of point cloud mapping are carried out on the TUM RGBD dataset,which proves the effectiveness of this method.