摘要
针对现有AD-Census代价计算时,Census变换依赖中心像素,对场景光照、亮度敏感的问题,提出一种基于子区域均匀度变换的改进代价计算方法。对Middlebury数据集在无幅度失真、不同亮度、不同光照场景下的代价计算以及立体匹配实验的统计结果表明,改进后的算法在非遮挡区以及所有区域的误匹配率均低于改进前,在代价计算环节,整体区域平均降低了5.80%,最终立体匹配后的视差图平均降低了1.55%。改进后的算法在不同亮度以及不同光照场景下对匹配的精度提升更加地显著,验证了改进算法的有效性。
Stereo matching is the key section in stereo vision,which includes cost computation,cost aggregation,disparity calculation and disparity optimization.The existing cost com⁃putation of AD-Census relies on the central pixel and is sensitive to scene lighting and brightness.Aiming at the problem,this paper proposes an improved cost computation method based on subregional uniformity.Datasets provided by Middlebury in three different scenes have been used in the experiment of cost com⁃putation and stereo matching.Compared to the AD-Census algorithm,the experimental results show that the improved al⁃gorithm reduces matching error in non-occluded areas and all areas.For cost computation and stereo matching,improved algorithm decreases matching error by an average of 5.80%and 1.55%in all areas.The proposed method enhances the matching accuracy more significantly when the stereo pictures are in different brightness or lighting,which verifies the effectiveness of the improved algorithm and its ability to deal with different scenes.
作者
胡璕
叶世榕
余振宝
黄亮
HU Xun;YE S hirong;YU Zhenbao;HUANG Liang(Guangzhou Urban Planning&Design Survey Research Institute,Guangzhou 510060,China;GNSS Research Center,Wuhan University,Wuhan 430079,China;China Academy of Civil Aviation Science and Technology,Beijing 100028,China)
出处
《测绘地理信息》
CSCD
2024年第1期138-142,共5页
Journal of Geomatics
基金
国家重点研发计划(2019YFC1509603)。
