Intelligent Retrieval of Surveillance Imagery

ObjectVideo (OV) proposes the development of tools and architecture for automatic forensic analysis and retrieval of video data. This effort enables accurate mining of surveillance video data based on target analysis and complex activity recognition. The video analytics solution proposed by ObjectVideo will entail a unified data management framework that integrates video, annotation text, maps and geo-reference data for robust video retrieval. OV’s existing Forensics product enables simple event retrieval based on target detection, tracking and classification algorithms. Under Phase I funding, we will develop algorithms to (a) recognize interactions between targets, (b) detect unusual and threatening multi-agent behaviors and (c) recognize targets in un-tagged data from examples. A powerful query/inferencing language will be used to allow users to formulate queries based on logical combination of target features, behavior descriptions and temporal constraints. Intuitive user interfaces will be incorporated for easy query creation and result visualization.