Person re-identification is the problem of recognizing people acrossdifferent images or videos with non-overlapping views. Although there has beenmuch progress in person re-identification over the last decade, it remains achallenging task because appearances of people can seem extremely differentacross diverse camera viewpoints and person poses. In this paper, we propose anovel framework for person re-identification by analyzing camera viewpoints andperson poses in a so-called Pose-aware Multi-shot Matching (PaMM), whichrobustly estimates people's poses and efficiently conducts multi-shot matchingbased on pose information. Experimental results using public personre-identification datasets show that the proposed methods outperformstate-of-the-art methods and are promising for person re-identification fromdiverse viewpoints and pose variances.
展开▼