In Partial Fulfillment of the Requirements for the Degree of Master of Science
will defend her thesis
Extracting Social Network Groups from Video Data Using Motion Similarity and Network Clustering
Detecting Social Networks from Video data acquired from surveillance cameras is a challenging problem currently being addressed by the Data Mining and Computer Vision Communities. As a part of continuing research in this area, a new graph-based post analysis approach is developed to process data obtained from the state-of-the-art Tracking and Identification systems to extract the various social groups present in it. The process of extracting social networks is primarily divided into two tasks. The first task consists of finding a method to compute a graph that connects all the people present in the video. Motion similarity between the tracks of the people on the ground plane is used as a metric to compute the weights on the edges of the graph. The second task is to cut the graph to form groups which is done by creating a minimal spanning tree and cutting the edges with least weights. The number of cuts to be made depends on the number of groups that are present in the video. To deal with the problem of unknown number of groups, the parameter of consistency of within cluster distances is exploited and the number of groups is decided by the finding the elbow point in the plot. The method shows promising results with UCLA Courtyard Dataset Videos and Simulation systems. This work can be regarded as one of the many approaches to solve the problem of “Detecting Social Networks from Video Data” which tend to exhibit decent outcomes.
Date: Monday, April 23, 2018
Time: 2:00 PM
Place: PGH 550
Advisor: Dr. Shishir Shah
Faculty, students, and the general public are invited.