Protocols for Evaluation of an Interactive Video Tracking System Utilizing Face Recognition

The Search and Retrieve prototype was developed as an interactive video tracking prototype to find and follow targets in a multi-camera surveillance system using face recognition. The overall potential benefits of using interactive face recognition for video tracking are unknown. We developed an evaluation protocol for the Search and Retrieve program using human-computer interaction and video tracking metrics. The protocol included three tracking cases: manual tracking, automated tracking using face recognition, and interactive tracking using both. We demonstrate that adding the operator's skill through interaction to the face recognition tracking present in the Search and Retrieve program can measurably increase recall by an average of 8 times for automated tracking and 39% over manual tracking in a limited time span of 20 minutes and without any prior training of the system for the user. The system's precision remains constant over the three cases.