The Research Of RGB-D SLAM System Based On Point And Line Feature

Simultaneous localization and mapping(SLAM)has been considered as a key technology for mobile robots to achieve autonomous navigation.Visual SLAM has been widely favored by researchers due to its ability of collecting abundant scene information.All the time,Visual SLAM has relaied on the point features in the scenes and ignored the rich line features in the structured environment.In this thesis,we use Kinect camera to construct a RGB-D SLAM system based on the point-line feature.The purpose is to improve the positioning accuracy and map reconstruction accuracy.The main contents of this thesis are as follows:Firstly,a visual SLAM system framework based on the Kinect camera was designed.The system was divided into two parts: front-end pose estimation and back-end pose optimization.The camera calibration experiment was performed on the Kinect by using the Zhang Zhengyou calibration method,registrating the RGB camera and the infrared camera according to the calibration results.Analyze the noise source of depth image and use bilateral filter algorithm to reduce noise.Secondly,the pose estimation algorithm based on the point-line feature is studied.The ORB point features and the LSD line features in the image are extracted,and the binary descriptor of the feature is calculated.Improve the accuracy and speed of mis-match culling by improving the RANSAC algorithm.The reprojection error model of point-line features and the analytic formula of Jacobian matrix are derived from the matching pairs of adjacent frames.Then,the pose optimization algorithm based on point-line features is studied.A key frame selection method is proposed,we can maintain the local map in real time and build a local map model to optimize the pose of the camera.Loopback detection is performed on the key frames,and construct pose graph to obtain a globally consistent motion trajectory,At the same time,the 3D point cloud map and octree map of the environment are reconstructed.Finally,a RGB-D SLAM system based on the Kinect and point and line feature is built for the research content of this paper.Multiple experiments are performed under the TUM dataset.It is verified that the proposed algorithm can improve the positioning accuracy and meet the real-time requirements.At the same time,it is performed in indoor scenes.Experiments verify that the proposed algorithm has a good accuracy of map reconstruction.