In the second stage of Chinese lunar exploration program, lunar rover will be launched for the moon. Stereovis ion system of the lunar rover is used for navigation and exploration. Binocular stereo vision cameras coUect stereoimage pairs of surrounding envrronment which are transmitted to ground station then used to reconstruct lunar 3Dterrain environment. 112 image pairs are collected at one site and terrain data increases rapidly as the rover exploresmore unknown area so that sequential process system is difficult to generate 3D terrain environment in time. Tosolve this problem, an elastic parallel co mputing platform is designed and built using OpenMP parallel co mputingframework on a 16-processor shared-memory machine. Key modules of stereo vision system are applied on theplatform. It processes image pairs and generates disperse 3D points cloud in parallel. Most important of all, theplatform can accelerate the speed of triangulation procedure for large dataset using elastic parallel Delaunaytriangulation algorithm designed in this paper. In this procedure, the platform elastically adjusts the number ofprocessors used in calculation to get the highest speed-up radio which can reach up t0 4.7 compared with sequentialsystem. The lunar rover stereo vision system has been tes ted using data from Chinese lunar terrain simulation ground.The experiment results show that elastic computing platform can obviously accelerate 3D lunar terrain generationspeed. To test performance of the system in several different size of dataset, we random generates datasets rangefrom 5,000 terrain points to 500,000 points. The elastic parallel computing platform only start several processors toprocess terrain data when dataset is small and add more processors as the dataset become larger. This paper presentsdesign and structure of elastic parallel computing platform based lunar rover stereo vision system. Additionally,paraUel Delaunay triangulation algorithm and its performance are described in details.
展开▼
