Stereo-based Dense Disparity Estimation For Lunar Surface

The lunar detection technology based on patrolling device instead of the human can observe at close range in lunar and implement scientific detection task, and is recognized as the most effective way on the detection of lunar and deep space.3D modeling system of stereo vision environment become the key technology of environmental perception for gaining direct, accurate and complete environment information and relative position between the patrolling device on the lunar surface.Above environmental background, the paper studies stereo vision dense matching algorithm mainly, and obtain high accuracy three dimensional image information of the lunar from the point of view to keep the technology consistency with the special lunar environment.Stereo vision dense matching algorithm based on the lunar surface is proposed in the stereo vision. The algorithm improved the effects of illumination through the calculation of intrinsic image and pretreatment of light consistency effectively. Improved BP algorithm that through being improved correlation calculation of MNCC pixels and using RGB triads for probability constraints of color similarity finished message propagation through parallel iteration of each direction carrying with the energy information of color similarity and smoothness, reduces the impact of discontinuous area and occlusion effectively. Mean Shift segmentation algorithm and adaptive threshold optimization which is adaptive dynamically to meet mean ratio of each pixel point to the all disparity value of area, are utilized to improve the precision of initial dense disparity map of weak and high similarity texture region in stereo images of the lunar.In3D reconstruction process, the paper introduces the calibration principle simply and gets internal reference and external reference of the image using simple, fast and accurate GML camera calibration software. According to the camera calibration parameters, epipolar is gotten to reduce the time complexity of matching algorithm greatly. Combining with the calibration parameters and fine disparity value to calculate3D coordinate values of each pixel, three-dimensional topographic map of the lunar surface is gotten by using OpenGL on texture mapping finally. The experiment results show that the algorithm is not only efficient in the running time, but also meet the special lunar surface environment and indoor and outdoor environment in the recovery effect, and confirms efficiency and robustness of the proposed algorithm profitably.