GE Global Research
Thursday, September 27, 2012
Title: Advanced Video Surveillance and Group Scenario Recognition
In this talk I will present research works from the Computer Vision Lab at GE Global Research, and focus on an integrated work on group-level video scenario recognition. The goal is to increase the situational awareness in public space and correctional settings, based on reliable detection and tracking of disorderly conducts and criminal behaviors such as fights, gang activities, and riots and provide timely alerts. Our system builds upon a real-time multi-camera multi-target tracking system. I will present two different methods that we investigated to detect and recognize complex and semantically meaningful behavior patterns. The first method is based on a probabilistic group structure analysis which evaluates the pairwise spatial-temporal tracking information. A path-based grouping scheme determines a soft segmentation of groups and produces a weighted connection graph where its edges express the probability of individuals belonging to a group. We derive probabilistic models to analyze individual track motion as well as group interactions. We show that the soft grouping can be combined with motion analysis elegantly to perform a large number of low- and high-level behavior recognition tasks. The second method is based on supervised machine learning to detect pre-defined scenarios. We extract features that capture the motion and action contexts among the groups, represented using the “bag-of-words” scheme. Group-level scenario context can be learned and a Support Vector Machine (SVM) is trained to classify a video segment into pre-defined event categories. Lastly, I will briefly present a multi-view, multi-target gaze tracking system that can operate with the behavior tracking system, capable to recognize attention and interaction among individuals.
Dr. Ming-Ching Chang is a computer scientist at the GE Global Research. He has conducted research on across a range of projects in computer vision and medical imaging, including intelligent video analysis for surveillance and tracking and modeling 3D shapes for visualization. He was a research assistant in the Laboratory for Engineering Man/Machine Systems (LEMS) at Brown University, where he received his Ph.D. in 2008. He was a research assistant in the Mechanical Industry Research Laboratories, Industrial Technology Research Institute (Taiwan). He received his M.S. degree in computer science and information engineering in 1998 and the B.S. degree in 1996, both from National Taiwan University. Dr. Chang has authored more than 25 peer-reviewed journal and conference publications and has filed 5 U.S. patent applications. He frequently serves as program committee and a reviewer for mainstream journals and conferences. He received the Green Belt Lean Six Sigma methodologies and Design for Six Sigma from GE in 2009.
GE Global Research