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.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
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)
关键词
双目立体测量
多视角深度图
配准
SIFT算法
特征匹配
binocular stereo vision
multi-view range images
registration
SIFT algorithm
feature matching