TitleVideo Surveillance Autopilot
Publication TypeConference Proceedings
Year of Conference2015
AuthorsSmith, LN, Bonanno, D, Doster, T, Aha, DW
Conference NameIEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2015
Date PublishedJune 2015
Conference LocationBoston, MA

Navy security watchstanders can be overwhelmed with the task of monitoring the content of multiple video streams for rare or important events. For example, if stationed on a ship in a busy port location, identifying those activities that require additional attention may benefit from familiarity with normal port operations. However, many activities may take place nearby (e.g., refueling operations, water taxi transportation, and leisure activities), and the constant monitoring of videos (and other sensors) can be taxing. We describe our plans for developing a Video Surveillance Autopilot (VSA), a software tool for automating surveillance tasks by providing watchstanders with tailored notifications and summaries of video content. Tremendous progress in deep learning and cognitive systems are making it possible to create such a system. We describe our vision of a VSA system. Given a video stream, the VSA (Figure 1) will consist of three primary components: (1) a text Annotator, (2) a Scene Interpreter, and (3) a User Interface (UI). Their developments pose substantial research challenges.

NRL Publication Release Number: