A System for Real-Time Recognition and Fast Retrieval of Close Contacts in Multi-Camera Videos

In recent years, the global community has been grappling with the COVID-19 pandemic. Rapid recognition and retrieval of infected individuals' close contacts could help mitigate the spread of the virus. The task is non-trivial for large scale video due to the tradeoff between effectiveness and efficiency. In this study, we develop a two-stage system that (1) automatically recognizes close contacts in real-time from video footage using multi-object tracking, and (2) fast retrieves the close contacts of a person queried by users in multi-camera videos through an Intra-Camera Association (ICA) approach. We also propose an efficient ICA method based on sparse trajectory correlation graphs. Experimental results demonstrate that our approach exhibits strong performance in terms of both effectiveness and efficiency.