Video ingress system for surveillance video querying

Abstract

Police departments struggle to review surveillance footage in an efficient manner. Finding a person matching certain characteristics requires manually scrolling through video feeds, potentially wasting hundreds of valuable man-hours solving crimes. SurvQ is a video query system which automates this process. Many existing systems are either too computationally intensive or only work on one type of camera. SurvQ, instead, focuses on a real time ingest system which supports arbitrary video sources (body cameras, dash cams, CCTV) at scale. It then uses a combination of cheap object detection, on-demand analysis, and priority ranking to efficiently analyze relevant video. We found this approach could scale to several hundred cameras in real time, suitable for the use-case of an entire police department in West Lafayette, IN.