PhD Prospectus Defense: Gregory Castanon
- 2:00 pm on Friday, December 20, 2013
- 4:00 pm on Friday, December 20, 2013
- 8 Saint Mary's Street - Room 404/428
Lighting Optimization with Interpretation of User Activity in Smart Lighting Environment Activity recognition in video is an emerging field in electrical engineering. With the affordability and popularity of video cameras, the quantity of video data available is growing faster than our ability to process or understand it. Towards the end of empowering users to navigate and reason over the content of these videos, we propose a framework for exploratory search over a large corpus of video data. Exploratory search is a user-driven process wherein a person provides a system with a query describing his interest. Typically, this description takes the implicit form of one or more exemplar videos, but it can also involve an explicit description. The system returns candidate matches, followed by query refinement and iteration. Success is judged by a precision/recall curve, as well as the run-time of the process. These two metrics yield a natural grouping of the challenges involved in exploratory search: quality of result, and speed of acquisition. Because many video systems continually stream in data, we seek in this prospectus to develop an archive which scales linearly with the data set and is calculable in real-time. In order to minimize user effort, we will create a search function that first efficiently downsamples the data to the set of potentially relevant vocabulary elements, enabling an efficient graph-based search. Subject to the constraints that archival processing should be feasible and real-time and search should happen in seconds, we aim to optimize the quality of the matches returned.