The aim of this project is to perform the automatic detection and characterization of events of interest (animal motion) in deep-sea videos and then to use the detected events as the basis for creating video abstracts. It is extremely time consuming for researchers looking for particular events of interest to manually search in the video database, and therefore, our study focuses on automatic detection of these events.

The experimental database consists of underwater videos from the NEPTUNE Canada archive. The NEPTUNE Canada cabled observatory is located off the west coast of Vancouver Island,  British Columbia. The network gathers data from undersea  instruments at depths ranging from 17 to 2,660 meters. The  instruments are connected to the Internet by fibre optic communication  cables that stream data to shore. All the data  gathered by the undersea instruments are transmitted to a data  archival system located at the University of Victoria, and are  openly available for download through a web-based interface.

Team:

• Alexandra Branzan Albu
• Aleya Gebali (M.A.Sc)
• Maia Hoeberechts (PhD), Research Coordinator, Neptune Canada

Sponsors:

• Neptune Canada
• IBM Streams

Publications

• A. Gebali, A. Branzan Albu, and M. Hoeberechts, 'Detection of Salient Events in Large Datasets of Underwater Video', in Proc. IEEE Oceans 2012, Virgina Beach, VA, Oct 14-19 2012.
• A. Gebali, 'Detection of Salient Events in Large Datasets of Underwater Video', M.A.Sc Thesis, University of Victoria, 2012.

Activity versus inactivity in sample video