引入垂直全局表观约束的中层特征行人再识别

发布时间：2018-04-19 01:21

本文选题：行人再识别 + 中层特征　；参考：《电信科学》2017年04期

【摘要】：行人再识别问题中,包含语义信息的中层特征能够提供更强的判别力。由于中层特征也采用局部匹配方式,与底层特征一样存在由于不同行人部分表观区域比较相似而产生误匹配问题。考虑到行人几乎都处于站立姿态,同一行人在垂直方向上的表观序列比不同行人的更相似,提出了在中层特征的基础上引入行人垂直全局表观约束,并融合底层稠密块匹配的识别方法。实验结果表明,算法在最具挑战的公用VIPe R数据库和CUHK01数据库上,均取得了比现有方法更高的命中率。
[Abstract]:In the problem of pedestrian rerecognition, middle level features with semantic information can provide stronger discriminant power.Because the middle level features also adopt the local matching method, the same as the bottom features, there is a mismatch problem due to the similarity of different pedestrian apparent areas.Considering that almost all pedestrians are standing, and the apparent sequence of the same person in the vertical direction is more similar than that of different pedestrians, the vertical global apparent constraint of pedestrian is introduced based on the mid-level characteristics.And fusion of the bottom dense block matching recognition method.Experimental results show that the algorithm achieves a higher hit ratio than the existing methods on the most challenging common VIPe R database and CUHK01 database.
【作者单位】：宁波大学信息科学与工程学院;
【基金】：国家自然科学基金资助项目(No.61175026) 浙江省自然科学基金资助项目(No.LY17F030002) “信息与通信工程”浙江省重中之重学科开放基金资助项目(No.xkxl1516,No.xkxl1521) 宁波大学胡岚博士基金资助项目(No.ZX2013000319)~~
【分类号】：TP391.41;TN948.6

【相似文献】