Detection of Fights in Crowd Video

Detection of fights and anomalous behavior of individuals in a crowd is a common problem in computer vision. Some tools that currently exist rely on optical flow of tracked features is a sequence of video frames. These motion algorithms are sensitive to independently moving objects in the frame. What constitutes an “anomaly” is context (eg. location) specific, thereby adding to the complexity. We aim to expand on existing work to create a baseline algorithm which can be used in a general use case, while in parallel investigate the state of the art; and if time permits, tune the algorithm to a specific use case

Suprit Singh
Faculty Supervisor: 
Viqar Husain
New Brunswick
Partner University: