Stereo Visual SLAM Based On Point And Line Feature

Visual Simultaneous Localization and Mapping(vSLAM)is an important research branch in the field of robotics,which has a wide range of applications.The traditional feature-based visual SLAM method relies on features in the image,and estimates the camera's motion trajectory by feature matching and construct an environmental map.However,the feature extraction and matching depend on the environment,and it's vulnerable to image noise,the mismatches in feature matching will affect the accuracy of pose estimation.In this paper,from the aspect of screening the camera with high precision feature matching for camera pose estimation,a stereo visual odometry combining the epipolar constraint and geometric information is proposed.The main research work includes:1.The epipolar and geometric information constraint are introduced into the features matching algorithm,and a line features matching method combining these constraints is proposed to reduce the cost of computing descriptor and improve efficiency.2.A feature robustness evaluation method is propoesd by research on the correlation among feature disparity and triangulation accuracy and feature matching accuracy.According to the result of feature robustness evaluation,this paper proposes an estimating poses method combining ICP and PnP,it improves the accuracy of pose estimation.3.A keyframe selection strategy is proposed to optimize the pose by local mapping.The method proposed is tested by KITTI public dataset,and compares with the visual odometry method PL-StVO,which is also based on the integrated point and line features,the effectiveness of the proposed method in estimating camera pose is verified.