Event Details
Detection of Salient Events in Large Datasets of Underwater Video
Presenter: Aleya Gebali
Supervisor: Dr. Branzan Albu
Date: Thu, June 9, 2011
Time: 10:00:00 - 11:00:00
Place: ECS 123
ABSTRACT
Abstract:
NEPTUNE Canada possess a large collection of video data for monitoring marine life. Such video data is of high interest for marine biologists who are able to observe species in their natural habitat on a 24/7 time frame. However, it is highly impractical for these researchers to manually locate the events of interest in such a large database. Our study aims to preform the automatic detection of the events of interest defined as animal motion. The proposed approach is based on optical flow techniques, which measure the displacement of pixels belonging to objects in motion on a frame-by-frame basis. This displacement is a vector that describes the amplitude and orientation of the velocity field. We adopt the Lucas-Kanade method due to its ability to track motion on a multi-scale basis, and hence to focus on large objects (such as fish) instead of small ones (such as marine snow).