Research On SLAM Algorithm Based On Binocular Visual Inertia

Autonomous navigation of mobile robots in unknown environments is one of the key technologies to make robots intelligent.Simultaneous Localization and Mapping(SLAM)is the core of autonomous navigation.Traditional visual SLAM will cause the loss of feature points and low positioning accuracy in the positioning system in the case that the overlapping area between adjacent frames is too small,the camera moves too fast and the image blurring,etc.Moreover,there are initialization scale problems and tracking scale drift problems in monocular SLAM.To solve the above problems,this paper focuses on SLAM algorithm based on binocular vision inertial navigation system.Firstly,for the traditional optical flow can not provide reliable initial estimation point and the feature matching accuracy is not high,a feature matching method combining binocular and Inertial Measurement Unit(IMU)information is proposed.The measurement point of the IMU is used to rotate the point to be tracked,and the point after the rotation is used as an initial estimation point of the optical flow,which improves the credibility of the initial feature point.In order to get accurate feature matching,the feature points are filtered and eliminated by using binocular camera baseline constraints and ring matching on the basis of optical flow tracking,and new feature points are extracted after each filter to ensure that there are enough feature points for each optical flow tracking.It is verified that after adding IMU and binocular information,the time required for optical flow tracking is shorter,and the accuracy of feature matching is higher.Secondly,a SLAM algorithm based on stereo vision and IMU fusion is proposed to solve the problems of motion blurring occurs when the camera is moving fast,too few overlapping areas between adjacent frames make it impossible to solve the pose,etc.An objective function consisting of visual reprojection error and IMU pre-integration residuals is constructed.In order to reduce the computational complexity,a sliding window method is used to ensure the real-time performance of the system,and the marginalized information is added to the objective function as a priori.Graph optimization method is used to optimize pose,speed,camera and IMU parameters.The proposed algorithm is validated in EuRoC dataset and actual laboratory scene.The experimental results show that the proposed algorithm can achieve localization and mapping in complex environments such as fast camera movement and light change,and the positioning accuracy is higher than that of open source algorithm Visual-Inertial System(VINS).Finally,for the problem that the selected feature points in the sparse map are the points where the pixels have significant changes,which can not represent the majority of the pixel depth in the entire scene,a dense map construction method based on binocular vision is proposed.The depth value calculated by the semi-global block matching(SGBM)is used as the initial value of the depth filter to update the depth,which improves the convergence speed.The Truncated Signed Distance Function(TSDF)method is used to fuse the depth after updating and the update frame by the depth after updating.The update frame is fused by the depth map acquired by binoculars,and finally a dense map is obtained.The algorithm is compared with the depth map and point cloud reconstruction effect estimated by REMODE.The results show that the proposed algorithm can get more depth maps,the time required for the image to be estimated is less,and the number of updated frames is less.At the same time,the TSDF method is used to construct the dense map of the EuROC dataset scene and some actual scene.