ECE Colloquium with James W. Davis

Clustering Using Ripley's K-Function for Analysis of Scene Behaviors in VideoProfessor James W. DavisDepartment of Computer Science and EngineeringOhio State UniversityRefreshments will be served outside Room 339 at 3:45 p.m. Faculty Host: Janusz KonradAbstract: In this talk, Professor Davis will present his research team’s work on an improved method for computing similarities of low-level video features which can then be clustered to extract various relationships among activity patterns in video for use in automatic visual surveillance tasks. He will focus on employing popular graph-based clustering algorithms, where performance depends heavily on the quality of the similarity matrix being clustered, which itself is highly dependent on pairwise scaling parameters. He and his team propose a novel technique for finding the scaling parameters using Ripley’s K-function to form local density-based neighborhoods based on random distributions. Additionally, Davis will provide a related hierarchical cluster merging approach to combine (grow) clusters in a bottom-up fashion. Results of the combined approach are first presented with synthetic datasets and then demonstrated with real video from surveillance cameras to extract behavioral patterns that identify popular pathway regions, detect anomalies, and retrieve similar video clips. Overall, he will show that multiple interesting and useful video understanding tasks can be accomplished via the clustering of low-level visual features.About the Speaker:James W. Davis is an Associate Professor of Computer Science and Engineering at Ohio State University. He received his PhD from the Massachusetts Institute of Technology in 2000. His research specializes in computer vision approaches to video surveillance and human activity analysis.

When 4:00 pm on Wednesday, October 31, 2012
Location Photonics Center, 8 Saint Mary’s St., Room 339