基于SIFT图像特征匹配的多视角深度图配准算法  

An Algorithm on Registration of Multi-view Range Images Based on SIFT Feature Matching

作　　者：韦虎[1] 张丽艳[1] 刘胜兰[1] 石春琴[1] 

机构地区：[1]南京航空航天大学机电学院,南京210016

出　　处：《计算机辅助设计与图形学学报》2010年第4期654-661,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(50875130);高等学校博士点专项科研基金(200802870016);江苏省科技支撑计划(BE2008136)

摘　　要：为有效地解决多视角深度图配准问题,提出一种新的配准算法.首先给出一种深度图数据图像化方法,根据深度图包含的像素信息和网格顶点处的曲率值创建特征图像;然后通过对特征图像进行SIFT特征检测与匹配来获得特征点与匹配关系,从而得到原始深度图上的特征点与匹配关系;最后采用投票和预配准方法去除误匹配,实现递增式多视角深度图配准.模拟噪声实验和多个实际测量深度图的配准实验结果验证了该算法的鲁棒性和有效性.A novel algorithm on registration of multi-view range images is proposed. Firstly, a method for transferring a range image to a feature image is performed, where the pixel information and the curvature information on the range image is used. Secondly, the keypoints and their matching relationships on the feature images are constructed by using the SIFT algorithm. The matching relationships among the original range images are obtained accordingly. Lastly, false matches are deleted by a voting and pre-registering scheme, and then multi-view range images are registered incrementally into a common frame according to the correct matches. Experimental results on both synthetic and real models demonstrate that the algorithm is robust and effective.

关 键 词：双目立体测量 多视角深度图 配准 SIFT算法 特征匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]