摘要
为了解决局部匹配算法误匹配率高的问题,提出一种基于AD-Census变换和多扫描线优化的半全局匹配算法。首先,通过绝对差AD算法与Census变换相结合作为相似性度量函数计算初始匹配代价,并构建动态交叉域聚合匹配代价;然后在代价聚合计算阶段,将一维动态规划的代价聚合推广到多扫描线优化,利用上下左右四个方向逐次扫描进行匹配代价聚合的计算,并引入正则化约束以确保匹配代价聚合的一致性,大大减少初始代价中的匹配异常点;最后,运用简单高效的胜者为王策略选出像素点在代价聚合最小时对应的视差,并在视差细化阶段,采用左右一致性检测和抛物线拟合方法进行后续处理以提高立体匹配的正确率。实验结果证明,该算法可获得高匹配率的视差图并且耗时较少。
In order to solve the problem of high mismatch rate of local matching algorithm,a semi-global matching algorithm based on AD-Census transform and multi-scan line optimization is proposed.Firstly,the initial matching cost is calculated by combining the absolute difference AD algorithm with the Census transform as the similarity measure function.At the same time,building dynamic cross-domain is to aggregate matching cost.Secondly,matching cost aggregating for one-dimensional dynamic programming is extended to the multi-scan line optimization.The calculation of matching cost aggregation is performed by successive scanning in four directions of up,down,left,and right,and introducing a regularization constraint algorithm is to ensure consistency of matching cost aggregation,which can reduce the a large number of matching abnormal points of the initial matching cost in the cost aggregation calculation stage.Finally,the Winner-Takes-All algorithm is used to select the parallax corresponding to the pixel when the cost aggregation is minimal,and the left and right consistency detection algorithm and parabola fitting algorithm are used for subsequent processing to improve the correct rate of stereo matching in the parallax refinement stage.Experimental results show that the algorithm proposed in this paper can obtain dense disparity map with high matching rate and less time consuming.
作者
朱荣华
葛广英
张广世
申哲
孙群
ZHU Rong-hua;GE Guang-ying;ZHANG Guang-shi;SHEN Zhe;SUN Qun(School of physics Science and Information Engineering,Liaocheng University,Liaocheng 252000,China;School of Mechanical and Automotive Engineering,Liaocheng University,Liaocheng 252000,China)
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2019年第9期978-984,共7页
Journal of Optoelectronics·Laser
基金
中央引导地方科技发展专项资金计划资助项目
