侯建华,周 浪,项 俊.基于边缘概率和m-Best求解的行人再识别[J].中南民族大学学报自然科学版,2017,(2):73-78
基于边缘概率和m-Best求解的行人再识别
Person Re-Identification Based on Marginal Probability and m-Best Solutions
  
DOI:
中文关键词: 行人再识别  图匹配  联合匹配空间  边缘概率分布
英文关键词: person re-identification  graph matching  joint matching space  marginal probability distribution
基金项目:国家自然科学基金资助项目(61671484);中南民族大学中央高校基本科研业务费专项资金(CZQ17001)
作者单位
侯建华,周 浪,项 俊 中南民族大学 电子信息工程学院武汉 430074 
摘要点击次数: 35
全文下载次数: 49
中文摘要:
      指出了行人再识别应用之一是在同一场景、不同摄像头视角下,对查询集与候选集中的所有目标做出最优关联,其本质是二分图匹配问题.针对传统方法只在联合概率分布基础上寻找一个最优解, 没有利用其它候选解中的有用信息,不能保证最优解的正确性的问题,提出了利用边缘分布特点, 综合其它候选匹配点的信息,在联合匹配空间上使每一对匹配点的边缘分布最大化,从而提高求解质量; 且对联合匹配空间的边缘分布计算不可行,采用二叉树分割算法求m个最优解,以实现边缘概率的估计.实验表明: 将上述方法应用于行人再识别问题能有效改善再识别算法的精度
英文摘要:
      One application of the person re-identification is optimal matching of all objects among prob set and gallery set which are under different camera views and at the same scene,and it is a binary graph matching problem in essence.The traditional method is to find a global optimum based on joint probability distribution that exploits no information of other candidates,thus can't guarantee the correctness even the obtained solution is optimal. To tackle this issue, an approach is proposed to maximize the marginal distribution for matching each pair points over joint matching space, and the matching solution is improved by taking into account all possible matching combinations for all other candidates. As exact marginalization of the joint matching space is intractable, binary partition algorithm is used to approximate the marginal distributions by exploiting the m-best solutions of the original problem. Experimental results demonstrate that the proposed method has improved the algorithm's accuracy when applied in the context of person re-identification
查看全文   查看/发表评论  下载PDF阅读器
关闭