An environmental change detection and analysis tool using terrestrial video

Abstract

We developed a prototype system to detect and flag changes between pairs of geo-tagged videos of the same scene with similar camera trajectories. The purpose of the system is to help human video analysts detect threats within a set of videos. While computers cannot differentiate threat from non-threat events, they can assist analysts by guiding their attention to sections of video where interesting events are more likely to appear. The system generates a single output video representing the difference between the input pair as well as a set of regions denoting sections of the world where changes occurred between the input videos. These regions represent segments of video where interesting events are likely to be seen. The difference video allows a video analyst to quickly see the differences between the two input videos and decide whether further analysis is required. The system is based on the video matching work by Seth Teller and Peter Sand [14].