摘要
本文提出一种面向流水线计算机的立体视觉并行匹配算法,使立体视觉算法所需的低层视觉信息处理和特征匹配都能在具有高速视频总线的流水线计算机中完成,这样既简化了视觉系统的结构,又大大提高了处理速度.匹配中应用了排序、方向和幅度约束作为相似性判断,并根据匹配点邻域中的视差梯度,利用松弛迭代法提高匹配的可靠性.算法已在PIPE流水线计算机上实现,256×256图像的立体视觉算法可在10秒内完成.
A parallel stereo matching algorithm has been proposed, which can perform both the low level information processing and feature matching of stereo vision by a pipeline computer with fast video bus. The ordering, direction and amplitude constraints are used as the similarity measure during the matching and a relaxation algorithm based on disparity gradient in the neighboring area of the match points is exploited to improve the reliability of the matching. The algorithm has been implemented on PIPE pipeline computer. The stereo algorithm of 256×256 resolution image can be finished in 10 seconds.
出处
《计算机学报》
EI
CSCD
北大核心
1993年第2期113-121,共9页
Chinese Journal of Computers
关键词
立体视觉
并行算法
面向流水线
Stereo vision, pipeline, parallel algorithm, match, relaxation.
