Person and Face Re-Identification Using Semantic Information and Single Shot Face Identification

Person re-identification is a contemporary problem involving the identification o f single person a cross multiple camera views, usually non-overlapping. The modern solutions of this problem involve deep learning based solutions, implicitly providing feature and metric learning. In this work we propose a novel approach for person and face re-identification guided by semantic information. Face re-identification i s approached using single shot face recognition for initialization using facial semantic parts. Person re-identification i s a lso guided b y semantic information represented by hands, legs, etc. It is further supported by the decision of face re-identification, if available. The proposed approach is tested on popular databases for person re-identification and demonstrate competitive results.