Research On Simultaneous Localization And Mapping Based On RGB-D Camera

Simultaneous Localization and Mapping(SLAM) could provide the means to make a mobile robot truly autonomous, at the same time,it is also one of the most challenging problems in the field of mobile robot. Visual SLAM technology has drawn attention of many researchers owing to its low price and rich information. Microsoft launched a RGB-D camera device in 2010,which cause a research upsurge in the field of SLAM because of its advantages mentioned above.SLAM could be generally divided into two dimensional SLAM and three dimensional SLAM with respect to the dimensionality of the map established, and their application domains are also different.Two dimensional map is commonly used for navigation and path planning of mobile robot, while three dimensional map could perform more advanced tasks, such as object recognition and so on. This paper mainly researched how to use low-cost Kinect to implement 2D SLAM and 3D SLAM.The character and calibration method of Kinect 2.0 were researched. The distance measurement principle of Kinect 2.0 was mainly studied,then the distinction between Kinect 1.0 and Kinect 2.0 was analyzed. Kinect 2.0 was actuated by Libfreenect2 in the Linux system and calibrated based on Zhang Zhengyou calibration method,thus intrinsic parameter and extrinsic parameter were obtained. Three-dimensionally chromatic pointclouds were generated by combining RGB image with depth image.Fake LIDAR data generated by Kinect which was applied in 2D SLAM was researched. ROS system and relevant tools were studied. Aiming at the problem that the depth data which were acquired by scanning of 2D LIDAR are incomplete in the vertical direction, and the depth data by Kinect were transformed into fake LIDAR data using depthimage_to_laserscan toolkit in ROS. In order to solve the holes problem existing in the depth image of Kinect, a simple method of depth image inpainting based on median filter algorithm and the principle of connected domain was proposed, and the corresponding experiment verified its validity. At the same time, fake LIDAR data generated by Kinect were applied in the Gmapping and Hector SLAM techniques, and the experiment in Gazebo and real enviroment demonstrated that it is feasible that fake LIDAR data generated by Kinect can be used in 2D SLAM techniques.The application of Kinect 2.0 in 3D SLAM was researched. ORB algorithm which is fast was used in the feature extraction and matching phases.To prove the rapidity of ORB algorithm, a comparative experiment among SIFT、SURF and ORB was implemented. PROSAC algorithm which is faster than RANSAC was used to remove outliers, and the experiment demonstrated that combining the ORB algorithm with the PROSAC algorithm for feature extraction and matching is real-time. Aiming at the defects of the classical ICP algorithm, such as low efficiency and local extremum in iteration, an improved ICP algorithm was presented and contrast experiment demonstrated high efficiency of improved algorithm. Finally,a SLAM experiment based on real environment was implemented using the improved RGB-D SLAM algorithm, and the environmental map and the trajectory of Kinect were obtained.