2024年
8月
图 学 学 报
JOURNAL OF GRAPHICS
第40卷 第4期
AugustVol.40
2024No.4
多测度融合的立体匹配算法研究
亢宇欣， 谌贵辉， 邓 宇， 张军豪
(西南石油大学电气信息学院，四川 成都 610500)
摘要：针对多测度融合的立体匹配算法的测度选择问题，提出一种基于测度互补系数的
测度选择方法。通过该方法选择多种测度进行融合作为匹配代价，并使用改进的半全局算法进行代价聚合，实现多测度融合的立体匹配算法。首先定义互补系数，通过互补系数选择多种相似性测度进行融合作为匹配代价；然后，针对半全局代价聚合使用随机初始化视差图导致立体匹配效果较差的问题，使用基于SURF特征得到的视差作为初始视差进行半全局代价聚合；最后计算视差并优化视差得到最终视差图。实验表明，使用该测度选择方法可以选择互补特征，结合改进的半全局代价聚合方法可以提高立体匹配效果。 关
键
词：多测度融合；测度互补系数；半全局代价聚合；立体匹配
DOI：10.11996/JG.j.2095-302X.2024040711 文 章 编 号：2095-302X(2024)04-0711-07
中图分类号：TP 391 文献标识码：A
Research on Stereo Matching Algorithms Based on Multi-Measure Fusion
KANG Yu-xin, CHEN Gui-hui, DENG Yu, ZHANG Jun-hao
(School of Electrical Engineering and Information, Southwest Petroleum University, Chengdu Sichuan 610500, China)
Abstract: Aiming at the measure selection problem of stereo matching algorithm based on multi-measure fusion, a measure selection method based on measure complementarity coefficient is proposed. Using this method, the present study fuses multiple measures as the matching cost, and adopts the improved semi-global algorithm for cost aggregation to realize the stereo matching algorithm of multi-measures fusion. Firstly, the complementary coefficients are defined, and a variety of similarity measures are fused by the complementary coefficients as matching costs. Then, in attempt to solve the problem of poor stereo matching effect caused by semi-global cost aggregation using randomly initialized disparity map, we carried out the semi-global cost aggregation taking disparity based on SURF features as initial disparity. Finally, the disparities are calculated and optimized to obtain the final disparity map. The experimental results show that the complementary features can be selected by using the measure selection method, and the stereo matching effect can be improved by combining the improved semi-global cost aggregation method.
Keywords: multi-measure fusion; measure complementarity coefficient; semi-global cost aggregation; stereo matching
双目视觉模拟人类视觉原理，实现从二维图像获取三维信息。立体匹配是指双目视觉系统针对同一场景在不同角度拍摄到的2幅图像，在搜索图像
中找到参考图像对应点的过程。立体匹配的效果决定双目视觉所获取三维信息的精确度，是双目视觉的重要环节。
收稿日期：2024-02-01；定稿日期：2024-03-21
基金项目：四川省科技计划项目(2016GZ0107)；四川省教育厅重点项目(16ZA0065)；南充市市校科技战略合作项目(NC17SY4001) 第一作者：亢宇欣(1993?)，男，河北廊坊人，硕士研究生。主要研究方向为计算机视觉、人工智能。E-mail：284260916@qq.com
多测度融合的立体匹配算法研究
