Welcome to the SceneDB project!

We are developing new methods and infrastructure for video analytics of longitudinal high-resolution video streams, with application to the Ocean Observatories Initiative HD undersea video feed . Large-scale video feeds are increasingly common in a variety of domains, but a framework for advanced analysis, information extraction, and "situational awareness" at scale does not exist. Scene analysis, anomaly detection, background filtering, and change detection over time are some of the initial tasks we seek to support, in addition to core curation and co-registration.

Current Status

Direct processing of the video streams is hindered by long downloading times and hard disk and memory constraints. The nature of the stream recording (with pan/tilt/zoom camera) requires additional reorganization of the video frames to obtain sequential monitoring of individual scenes. We have developed a framework which allows users to query individual scenes across time. The scenes are extracted automatically, and stored in a database of scenes (SceneDB). The approach would facilitate both end users who wish to explore the videos and data analysts who would like to perform longitudinal analysis and test scientific hypotheses.

Workflow Diagram (click to enlarge).

Participants

Sponsors