The Research On Algorithm Of Unmanned Planet Rover Vision System

As human extend the exploration range of the outer space, people started to seek for the approaches of remote planet exploration. Nevertheless, unexplored planet exploration is a high risk action while the problem of manned long distance space traveling is unsettled. For all of these reasons, unmanned planet exploration rover can be sent as the pioneer of the exploration projects. The rover doesn't need life support system. It can fulfill most missions which formerly can only be done by human. In comparison with sending human to outer space planet directly, planet rover is more economic and have low risks.This thesis introduces the backgrounds of planet exploration rover, review the development of planet rover navigation system, and carefully research the main functions of vision system for planet rover autonomous navigation. The thesis start from the architecture of vision system and defined the basic element of vision system, the missions and core contents of vision system algorithms are discussed. The interests of this thesis are mainly focused on the two key points, stereo vision and motion estimation.The fundamental of stereo vision is triangulation, for this reason, precise camera geometry model and correspondence results are the most critical problems in this issue. First of all the basic idea of calibration and rectification are analyzed, after that, Zhang's calibration method has been discussed as an effective algorithm. Through the application of Zhang's calibration method, the cameras are well calibrated and rectified. Existing common methods of correspondence are carefully studied, gives a feasible correspondence method include three main steps:firstly, using DOG (Differential of Gaussian) as pre-process method; then SAD (Sum of Absolute Difference) for calculation of local correspondence costs; and after all disparity is refined by some methods. The theories of three dimension reconstruction are given subsequently. The results of correspondence and reconstruction are illustrated.This thesis also introduce an approach of motion estimation based on feature extraction technology, study some basic theory of scale invariant feature transform (SIFT). Optimize the realization of SIFT. Present a motion estimation method based on combination of landmark feature extraction and the results of three dimension reconstructions, through identify the feature of landmark and check the transformation of landmark point's space coordinates location to estimate the motion of rover. Finally, the measurement results are given, the results meet the requirements of precision.Based on the research of these two algorithms, this thesis gives a solution to the rover vision system algorithm. It is significant to the related research of rover navigation system.