Hello, my name is Sahil Shah and I am participating in a Research Experience for Undergraduates (REU) at the computer vision lab at the University of Central Florida.
My work for the summer involves Multimedia Event Detection, which is a fancy term for classifying videos based on the actions within it. For example, a video containing a baseball diamond and pitcher may be detected as "hitting a home run." The project is motivated by NIST'S TRECVID competition. Specifically, my part in the project is looking into how Hidden Markov Models and other probability models could be used to further improve classification performance by exploiting the sequential information embedded in video annotation data. The reasoning is that certain actions in a video occur sequentially, like how the pitcher throws the ball before it is hit by a batter. By incorporating temporal data, we add another tool to our repertoire for correcting classifying videos.
I collaborated closely with my graduate student mentors Subh Bhattacharya and Hamid Izadinia, as well as Dr. Mubarak Shah.
- TRECVID 2010(UCF 2010 submission paper)
- A Revealing Introduction to Hidden Markov Models
- Behavior Recognition via Sparse Spatio-Temporal Features
- Towards optimal bag-of-features for object categorization and semantic retrieval
- Distinctive Image Features from Scale-Invariant Keypoints
- On Space-Time Interest Points