Towards a Framework for Context-based Online Person Identification

Abstract

In this paper, we propose to use a combination of machine and deep learning tools for online person identification (PI) in a video-surveillance-based tracking system. To this end, we propose a skeleton of an algorithm-based framework that merges face and cloth-based identification to cope with the limitations of each one. We aim especially to complement face recognition based identification by clothing attributes based techniques by using contextual information to deal with complex conditions where there is a variability in lighting, pose, face size and distance from the camera. We therefore proposed to use context information to jointly integrate facial recognition and clothing recognition in unifying framework.