Person Re-Identification Using Group Constraint

The group refers to several pedestrians gathering together with a high motion collectiveness for a sustained period of time. The existing person re-identification (re-id) approaches focus on extracting individual appearance cues, but ignores the correlations of different persons in a group. In this paper, we propose a group-guided re-id method named group retrieval correlation (GRC) to address the above problem, which pays more attention to the correlations of surrounding pedestrians in the same group and reliefs the interference caused by the dependence of appearance cues. Compared with traditional re-id methods which compute naive similarity merely by the appearance characteristics, the GRC based re-id method proposes a novel and optimal person similarity by considering the relationships among groups. Therefore, the proposed approach provides sufficient hints of group relationships, which are supplementary to the appearance features and contributes to constructing a more reliable re-id system. Experimental results demonstrate that the group information promotes person re-id by a large margin on the proposed Group-reID dataset in terms of both hand-crafted descriptors and deep features.