Research On RGB-D SLAM Algorithm Based On Kinect

SLAM solves the problem of how the machine senses the environment in an unknown environment.Kinect is widely used in visual SLAM because it can directly acquire depth information and save valuable computing resources.This paper mainly studies the visual SLAM algorithm based on Kinect camera from the following four aspects:Firstly,the imaging model of the Kinect camera is analyzed to complete the calibration experiment of the color camera and the depth camera of the Kinect camera to ensure the accuracy of the image data.Then,the related algorithms of feature detection,extraction and matching in the front-end part are analyzed and compared,and the algorithm of feature matching and culling mismatching is improved,so that more correct matching pairs can be used for pose calculation.In order to solve the pose calculation failure caused by the lack of depth,ICP and PnP is used to estimate the camera pose.Secondly,the Loop-closure-detection based on key frame and the bag-of-word model is added to the backend optimization module,which reduces the cumulative error generated by the front end.Loop-closure-detection adds constraint edges to the pose model according to the interframe relationship of the front end,and enhances the system.The global consistency makes the system maintain high precision during long-term operation and the system robustness is improved.Finally,the algorithm implementation and experimental verification of the algorithm are based on the software algorithm implementation platform of Linux system and the mobile robot hardware platform based on Kobuki,then the verification experiment with TUM data set,aiming at the size and texture details of the indoor space.The comparison between the two aspects of size of experiment space,analysis of positioning accuracy and mapping effects,compare the efficiency of the three SLAM algorithms in the same scene.Field experiments were conducted in a laboratory environment.