摘要
视频中行人有时会相互交错,导致行人被部分或全部遮挡。针对视频中遮挡行人再识别问题,提出了一种基于人体姿态估计算法提取局部特征的行人再识别方法。与以往使用的贴片、条纹等提取的局部特征包含大量噪声不同,该方法改用人体姿态估计算法提取更精确的行人刚体部分;通过计算刚体部分的颜色直方图及其颜色直方图综合相似性得分,选取视频帧中得分靠前的候选行人;采用马氏距离代替传统的欧氏距离分别对候选行人进行距离度量。实验结果表明,所提出的算法简单、快速,可以有效的用于视频中遮挡行人的再识别。
Pedestrians sometimes interlace in the video and are partially or completely obscured.To solve the problem of occult pedestrian re-identification in video,a pedestrian re-identification method based on local features is proposed.Compared with the patch used in the past,the extracted local features such as stripes and the like contain a large amount of noise,and the human body pose estimation algorithm is used to extract a more accurate pedestrian rigid body portion.By calculating the color histogram of the rigid body part and its color histogram comprehensive similarity score,the candidate pedestrians with the highest score in the video frame are selected.Finally,the distance measurement of candidate pedestrians is performed by using the Mahalanobis distance instead of the traditional Euclidean distance.The experimental results show that the proposed algorithm is simple and fast,and can be effectively used to occlude pedestrians in the video.
作者
魏英姿
杨继兰
WEI Yingzi;YANG Jilan(Shenyang Ligong University,Shenyang 110159,China)
出处
《沈阳理工大学学报》
CAS
2020年第1期49-53,94,共6页
Journal of Shenyang Ligong University
基金
辽宁省自然科学基金项目(20180550791、20180550520).
关键词
人体姿态估计
颜色直方图
马氏距离
行人再识别
human body posture estimation
color histogram
mahalanobis distance
pedestrian re-identification
