A Unified Framework for Face Re-Identification Using Entry-Exit Cameras

Robust re-identification of individuals across multiple cameras has become a fundamental requirement in modern security and surveillance. To address this, we propose an effective system that ensembles detection, tracking, feature extraction, and retrieval mechanisms into a unified pipeline. By leveraging lightweight models, the system minimizes computational overhead and functions effectively on surveillance cameras. The system architecture enables processing and reliable identification, making it suitable for real-time applications. Our approach involves logging entry/exit registries and matching exits with entries through a re-identification pipeline, which allows security personnel to monitor an individual's entry and exit times within a secured premise. We are also releasing an open-source test video dataset with annotations, which is helpful for evaluating the face re-identification approaches for entry-exit cameras. Our approach performed good while considering matching of exit and entry events. The registered persons were identified with much better accuracy considering entry/exit events.