People Tracking and Re-Identifying in Distributed Contexts: PoseTReID Framework and Dataset

We introduce a generic framework which is designed for effective real-time 2D multi-person tracking in distributed people interaction spaces like malls or amusement parks where long-term people's identities are important for other studies such as behavior analysis. The framework relies on multi-person pose detector for detecting bodies' parts and a recognizer for re-identifying people. We carefully selected existing sub-modules for our contexts, and the framework can efficiently track people and re-identify them using faces even later after tracking losses or reappearances of people. Along with this paper, since all existing datasets for people tracking barely have visible faces, we also introduce our people tracking dataset which is specifically designed for distributed people interaction spaces where people's faces are visible and recognizable. The results of the proposed PoseTReID framework are very interesting in all scenarios when compared on our dataset to a recent state-of-the-art tracking method. This efficient people tracking framework in distributed and interactive contexts, which is achieved here, is an important brick towards future works of people grouping and behavior analysis.