A spatial and temporal features mixture model with body parts for video-based person re-identification

摘要

The goal of video-based person re-identification is to recognize a person at different camera settings. Most previous methods use features from the full body to represent a person. In this paper, we propose a novel Spatial and Temporal Features Mixture Model (STFMM). Unlike previous approaches, our model first horizontally splits human body into N parts, which include the information of head, waist, legs and so on. The feature of each part is then integrated in order to achieve more expressive representation for each person. Experiments conducted on the iLIDS-VID and PRID-2011 datasets demonstrate that our approach outperforms the existing video-based person re-identification methods and significantly improves stability. Our model achieves a rank-1 CMC accuracy of 73.6% on the iLIDS-VID dataset and a rank-1 CMC accuracy of 47.8% for the cross-data testing.