基于双目视觉的地形三维重建  

3D Terrain Reconstruction Based on Binocular Vision

作　　者：周瑜[1] 李戈[1] 张学贺[1] 赵杰[1] 

机构地区：[1]哈尔滨工业大学机器人技术与系统国家重点实验室,黑龙江哈尔滨150001

出　　处：《机械与电子》2016年第7期57-61,66,共6页Machinery & Electronics

基　　金：国家自然科学基金资助项目(61473104)

摘　　要：为克服传统接触式测量方法的缺点,更便捷更准确地获取地形信息,提出了基于双目视觉的地形三维重建方法。该方法对双目图像对进行了校正;并采用高效率高准确度的支撑点邻域扩展立体匹配算法获取空间点的三维信息;最后基于八叉树原理,采用曲率法精简点云,并结合改进的Delaunay三角剖分算法进行三角网格化,从而恢复地形三维模型。实验结果表明,采用本算法能够在1s内获得致密的视差图,将21 841个点云精简到5 463个(25%)仅用时0.7s,并且能在此基础上高效地完成三角网格剖分,实现崎岖地形的三维重建同时保留地形细节特征。In order to conveniently obtain topographic information and overcome the draw back of the traditional contact measurement methods ,a 3D terrain reconstruction method based on binocular vision was put forward .Firstly ,image correction was completed .Next ,3D point information of terrain was ob‐tained through an efficient stereo matching algorithm based on supporting point neighbor extending .And then ,a curvature method based on octree principle was used to simplify the point cloud .Finally ,an im‐proved Delaunay triangle subdivision algorithm was used to accomplish the triangulation and rebuild the 3D model of terrain .Experimental results have shown that the proposed methods in this paper can acquire dense disparity map in 1 second and reduce the number of point from 21 841 to 5 463 （25% ） within 0 .7 second .And on this basis ,triangulation of the rough terrain for 3D model rebuilding can be quickly a‐chieved with the details reserved .

关 键 词：双目视觉 立体匹配 点云精简 三维重建 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]