Abnormal Crowd Behaviour Recognition in Surveillance Videos

The paper presents an approach to crowd behaviour recognition in surveillance videos. The approach is based on a 4-stage pipelined multi-person tracker adapted to microscopic crowd level representation and crowd behaviour recognition by the evaluation of fuzzy logic functions. The multi-person tracker combines a CNN-based detector and an optical flow-based tracker. The following tracker features are used: optical flow and histogram of optical flow orientation at the macroscopic level, and the tracklets and trajectories of a person and/or group of people at the microscopic level. The human interpretation of video sequences (real and/or video sequences obtained by simulators of crowds) is mapped into fuzzy logic predicates and fuzzy functions. Fuzzy logic predicates specify crowd motion patterns at the microscopic level for a person and/or group of people. They are building blocks of fuzzy logic functions which describe different scenarios of characteristic crowd behaviour. The preliminary results of three experiments for a runaway scenario show that the approach supports efficient and robust crowd behaviour recognition in surveillance videos.