Research On Binocular Stereo Vision System For Indoor Mobile Robot On Complex Environment

In Chapter 1, the research status of binocular stereo vision is described. General map building methods and vision-based obstacle detection methods are summarized. Binocular stereo vision systems for mobile robot navigation both at home and abroad are studied. Finally, the significance and main content of this research topic are discussed.In Chapter 2, the pinhole model of a single camera is introduced, also the lens radial distortion and tangential distortion is taken into consideration.2D planar checkerboard image is applied for camera calibration. Using Open Source Computer Vision Library (OpenCV), the intrinsic and extrinsic parameters of the single camera and the space parameter of the stereo vision system. Also, for the convenience of stereo matching, the image pairs are rectified using Bouquet method.In Chapter 3. the internationally popular stereo matching algorithm criteria and standard test images are present, which provides an objective basis for the evaluation of stereo matching algorithm. There are mainly two categories of matching algorithms, local and global matching algorithm. General matching algorithms of each category are studied, for example the Winner-Take-All, dynamic programming, and graph cut methods. Experiment results compare these methods quantitatively and qualitatively.In Chapter 4, according to the requirements of the mobile robot application, th local area-based matching algorithm is deeply researched. The accuracy of different similarity measure functions of and matching window size are analyzed. In order to improve the quality of disparity map, image pre-and post-processing methods are applied. Furthermore, the filter box and multi-resolution and other computing technologies to accelerate the implementation of stereo matching algorithm are used.In Chapter 5, the ideal model for the depth of binocular stereo vision system measurement is analyzed. The local method for map building method is studied, reasonable height model is used to remove surface noise. In this chapter, calibration results and stereo matching results are tested by depth measurement. The depth calculation results show that the depth measurement error of the actual screen will meet the needs of practical application.3D reconstruction and occupancy map results can provide enough environment information for the mobile robot, which will be the foundation for path planning.