| Along with rapid economy developments and successful launching of Shenzhouâ…¤and Shenzhou VI spacecrafts of China, deep space-exploration will become the next research goal. At present, Chang E project not only requires soft landing but also patrolling and reconnaissance on moon surface by unmanned exploration in 2012. So an autonomous navigation lunar rover should be developed to be adaptive to harsh environments of moon.Environment recognition based on binocular vision has some features such as compact structure,low energy consumption,rich important and high identifying precision. This system has been paid a great attention and already been applied to patrolling explorer to carry out navigation and obstacle avoidance.Intelligent Vehicle group of Jilin university and CAST has cooperated to develop a theory model of lunar rover. Binocular stereo vision system was built and off-line calibration was performed. Firstly stone's edge information is derived from two images of CCDs, then matching points of edge maximum points can be get in virtue of epipolar line. Secondly world coordinates and camera coordinates of these matching points can be calculated. Finally, 3-D information of the stone can be derived. This algorithm not only resolved overlapping and cuboids stone's flat top problems but also had real-time performance and high measuring accuracy.The title has been listed on the top of this abstract. The paper's main content includes five parts, they are as follows:(1): Construction of binocular vision system; (2): Calibration of binocular vision system; (3): Stereo matching of images; (4): epipolar constraint and fundamental matrix; (5): 3-D information obtaining of stones by epipolar constraint.The paper introduces theories of active vision system and passive vision system. Although active vision system has some advantages in vision field and flexibility, the found of this system is relatively complex and requires additional measure equipments. So the passive vision system is chosen because of its advantages such as simple system, less calculation, mature technology and extensive application. The same time various instrument and parameter are decided to found this system in view of requirements.The paper calibrated the system and determined extrinsic and intrinsic parameters of camera to set up corresponding relations between object points in world coordinate and image points in image coordinate correctly. Camera calibration generally includes off-line calibration based on referenced objects and self-calibration based on active vision. The paper studies both calibrations. Self-calibration has advantages of convenience and quickness in use, but it has great calculation and unstable value, so the off-line calibration is chosen because of its simple algorithms and stable calibration result. In order to get minimum calibration error, Least Mean Square method is used to calibrate the system and has a test of calibration result-projection matrix, it comes to the conclusion that the system has high calibration precision and can satisfy our study.Image matching is the most important and difficult step in binocular vision and there isn't a widely-used matching algorithm to this day. The paper summarizes a few of current representative algorithms including template matching, character string matching and feature points matching. And then compares their environment adaptive ability, advantages and shortcomings. In allusion to the need of lunar rover, the paper adopts a combinative algorithm which is based upon template matching and feature points matching. In the end some experiments are carried out on this algorithm and has a conclusion that it has high precision when image gray has some changes and in other cases it will result in mistake matching.The matching algorithm based upon template matching and feature points matching searches matching points along X and Y directions on the image plane, it's obvious that this process is two-dimensional. In order to accelerate matching, this paper adopts epipolar constraint to change matching process from Two-dimensional to One-dimensional. Epipolar constraint can be expressed as a three-multiply-three fundamental matrix. Least Mean Square method and RANSAC are used to compute fundamental matrix. Least Mean Square method is high in precision, but RANSAC is better has the higher precision in result and more stable result because it can reach to sub- pixel grade, so RANSAC is used to compute fundamental matrix in the paper .About the measurement of stone, the paper introduces the vir stone arithmetic of primary measurement and points out its disadvantages in course of measuring stone. This arithmetic can't resolve occlusion and cuboids stone flat top problems, so the paper put forward a new measurement -stone division measurement. At first we extract edges of stone in the image, and then we divide the edges of stone into four parts, they are top, bottom, left and right. Passivitye sentence should be written work out the world coordinate and camera coordinate of maximum points of four edges; finally Passivitye sentence should be written can get 3-D information of the stone. In order to resolve occlusion problem, the measuring of left edge is an example, passivitye sentence should be written the left edge of right image as datum because right image has occlusion problem. Then matching points are searched in the left image, meanwhile passivitye sentence should be written conventional rectangle template into triangle template and passivitye sentence should be written matching speed by means of epipolar constraint. Experiments prove that this arithmetic can quickly resolve occlusion problem and get high precision of width information of the stone. In view of the cuboids stone flat top problem, as Z axes of the world coordinate is vertical and upwards, so passivitye sentence should be written use Z value of the world coordinate between maximum points of top edge and maximum point of bottom edge to get the height of the stone. Experiments prove that this arithmetic can resolve cuboids stone flat top problem efficiently. The two methods to get 3-D information of stone are used in the paper, namely, the method of single point and the method of many points which are based on epipolar constraint. The method of single point chooses only one edge maximum point to carry out matching process in every edge, but the method of many points chooses many edge maximum points to carry out matching process in every edge, then passivitye sentence should be written noise points through error operation, in the end we choose a point in residual points for the last edge maximum point. Experiments prove that the method of single point is better than the method of multi points in the precision and the speed, so the method of single point based on epipolar constraint is used.The application software is developed through Visual C++. Experiments of stone are carried out and the experimental results prove the validity of the algorithms mentioned in this paper.
|
PDF Full Text Request