摘要
计算机双目视觉技术是自然光条件下可使用的非接触测量方法,对于逆向工程、工业检测、三维重建、移动机器人导航等领域都有重要的实际应用价值.针对动态规划立体匹配中存在的条纹瑕疵问题,本文提出一种新的动态规划立体匹配方法,设计了新型矩形窗口计算匹配代价,构建基于遮挡约束的优化函数,在降维低分辨率图像上求取控制点抑制条纹瑕疵.选择Middlebury平台上的4个组的图像作为实验图像,将基于置信传递的立体匹配方法和基于图形切割的立体匹配方法作为比较方法,对本文方法的性能进行验证.4个组的实验图像的立体匹配实验结果表明:基于控制点修正的动态规划立体匹配方法,获得的视差图像连续稠密、坏像素比例低,明显优于两种比较方法.应用本文方法,对比基于图形切割的立体匹配方法(GC)和基于置信传递的立体匹配方法(BP)获得的视差图像,坏像素比例分别下降了1~2个百分点.
Computer binocular vision technology is a non-contact measurement method that can be used under natural light conditions.It has important practical application value in reverse engineering,industrial detection,three-dimensional reconstruction,mobile robot navigation and other fields.In this paper,a stereo matching method based on dynamic programming optimization strategy has been proposed,and innovative work been carried out in matching window design and disparity calculation optimization.A rectangular window has been designed to calculate the matching cost,and a dynamic programming process based on occlusion constraints been constructed to obtain control points to suppress fringe defects on dimensionally reduced images.Four groups of images on Middlebury platform have been selected as experimental images.The stereo matching method based on confidence transfer and the stereo matching method based on graph cutting have been selected as comparison methods.The performance of this method is verified.The results of stereo matching experiments of four groups of experimental images show that the dynamic programming stereo matching method based on control point correction can obtain disparity images with dense continuity and low proportion of bad pixels,which is obviously superior to the two comparison methods.
作者
许翊
刘学军
XU Yi;LIU Xue-jun(Zhenjiang Higher Professional Training School, Danyang Normal College, Danyang Jiangsu 212300, China;College of Computer Science and Technology, Nanjing University of Technology, Nanjing Jiangsu 211816, China)
出处
《西南师范大学学报（自然科学版）》
CAS
北大核心
2020年第9期118-123,共6页
Journal of Southwest China Normal University(Natural Science Edition)
基金
国家重点研发计划项目(2018YFC0808505).
关键词
双目视觉
立体匹配
动态规划
视差图像
binocular vision
stereo matching
dynamic programming
parallax image
