Kitware is leading the development of a Video Image and Retrieval Analysis Tool (VIRAT). The VIRAT system features a revolutionary video analysis capability that filters and prioritizes massive amounts of archived and streaming video based on events; presents the high-value intelligence content clearly and intuitively to video analysts; and results in substantial reductions in analyst workload while increasing the quality and accuracy of intelligence yield. Focused on aerial surveillance video taken from Predators and other drones, VIRAT automatically extracts descriptors of events and humans in the video, emitting real-time alerts of events of interest. The system also indexes the descriptors into a database to enable subsequent search for similar and related events, with examples of distinguishable events ranging from single-person (walking, limping, gesturing, etc.) and person-to-person (meeting, following, gathering, etc.) behaviors to person-vehicle interactions, among others.

As improvements in technology allow drones and other tools to capture more video data, it is essential to provide analysts with tools that facilitate rapid, efficient analysis so they can identify, assess and act on image and video segments with noteworthy content or events.

In our CVPR 2011 paper we introduced a new dataset to assess the performance of continuous video event recognition algorithms. We believe that this will stimulate research in event recognition and other diverse areas of computer vision in the years ahead!