Research On The Key Technology Of Vision-based Navigation: Stereo Vision And Path Planning

Autonomous Land Vehicle (ALV) is an intelligent mobile robot, which can run autonomously and continuously on road or cross-country. Its research, which involves theories and technologies of many disciplines, and embodies the latest achievements of information and artificial intelligence, is of great value in research and application, and is paid great attention to all over the world. Of all the ALV key technologies, vision-based navigation perceives and understands the surrounding quickly, and then determines a negotiable region for the ALV. The key to the vision-based navigation is the rapid detection and recognition of obstacles. The basic strategy of path planning is that the ALV plans a safe and efficient path in the negotiable region based on the information from vision processing. Stereo vision gets the information ahead and the intelligent control technology makes the ALV move along the planned path. This thesis aims at the design and implementation of the stereo vision system as well as path planning based on the vision information.Perhaps the most important problem for stereo vision is the accurate calibration of the cameras. In chapter 2, the main advantages and disadvantages of various of calibration objects, control points and exiting camera models are compared and analyzed, then suitable ones for the ALV are determined, and the corresponding calibration method is discussed. This method only requires the camera to observe a planar pattern shown at a few different orientations. Either the camera or the planar pattern can be freely moved. The proposed method gains the flexibility to the work environment for the ALV, and very good results are obtained. Rectification is also a very important procedure to reduce the complexity of stereo matching. The rectification algorithm for binocular stereo vision is studied firstly. Then the above algorithm's limitation is analyzed that can not be applied in the rectification of multiple-baseline stereo vision images. The algorithm is modified and is used to rectify the parallel trinocular stereo vision system. The experimental results show that this algorithm can not only rectify the images for parallel trinocular stereo vision system, but also be extended to the parallel multiple-baseline stereo vision system with more than three cameras.In chapter 3, stereo matching algorithms are studied thoroughly and the emphasis is put on their speed and robustness. To area-based stereo matching, the topics of our research include choosing the matching criteria, window sizes, preprocessing and post-processing methods. The qualitative and quantitative analyses have been made on the matching results of the synthesized and the real stereo pairs. And various of accelerating techniques, including multi-resolution, box-filtering, parallel instructions, hyper-threading, OpenMP are used to shorten the runtime. To global stereo matching, the popular algorithms, dynamic programming and graph cuts, are evaluated qualitative and quantitative. In order to overcome the time-consuming of above algorithms, a multi-resolution-based matching method is proposed and much time is saved.The results of stereo matching are expressed by 3D reconstruction. In chapter 4,a real-time 3D reconstruction algorithm is proposed. Combined with various of error sources which come from the model for calibration, image noise, foreshortening error,misalignment error, system error, discretization and the binocular stereo vision, the errors of reconstruction are analyzed. Then the method to compensate the errors is proposed and proved to improve the precision. Using the 3D reconstruction algorithm, we build the global map for cross-country based on the information from the GPS/INS and the map almost reflects the real scene. The ALV can move safely with the help of the information from 3D reconstruction as well as corresponding path planning which produces a safe, efficient and collision avoidance path. In order to overcome the drawbacks of traditional path planning algorithms, i.e. poor adaptability to env...