PhD Proposal: Improving Detection in Computer Vision

Extracting information automatically from large databases of images and videos is central to many problems in computer vision. In general, detection is the extraction of particular information from a larger stream of information without specific cooperation from the sender. To this end, we propose two different approaches for detecting "actions", "activities" and "events" in videos. For performing temporal localization of actions and activities in videos, we propose a multi-stream bi-directional recurrent neural network based architecture, which captures both short and long term temporal dependencies. In another setting for event detection, when no training data is provided a-priori, we propose an algorithm that constructs pairs of automatically discovered concepts and then prunes those concepts that are unlikely to be helpful for detection.

Examining Committee:

Chair: Dr. Larry S. Davis

Dept rep: Dr. Tom Goldstein

Member: Dr. David Jacobs