Semi-Interactive Tracing of Persons in Real-Life Surveillance Data

To increase public safety, more and more surveillance cameras have been placed over the years. To deal with the resulting information overload many methods have been deployed, focusing either on real-time crime detection or post-incident investigation. In this paper we concentrate on post-incident investigation i.e. crime reconstruction using video data. For a complete crime reconstruction, the location of all persons of interest should be known before and during the incident. To do so, we follow persons within the field of view of a single camera (tracking) and between different cameras (tracing).

We present a semi-interactive approach to post-incident investigation. This method is specifically capable of tracking and tracing persons of interest. Our system supports the analytical reasoning process of the investigator with automatic analysis, visualization methods, and interaction processing. We show that the automatic tracing method significantly speeds up tracing of persons with clear visual characteristics. Tracing of persons without obvious characteristics is an inherently difficult task, but we show that intelligent use of interactive methods greatly improves the tracing performance of our system.