Abnormal crowd motion detection with hidden conditional random fields model

Abstract

Crowd motion analysis in public places is an important research subject in the monitoring field. This paper proposes an approach for detecting abnormal crowd motion using Hidden Conditional Random Fields Model (HCRF). This approach derives variations of motion patterns from direction distribution of the crowd motion obtained by the optical flow and these variations are encoded with HCRF to allow for the detection of abnormal crowd motion. Modeling the temporal neighborhood relations in a video sequence based on HCRF can incorporate hidden states and label the video depending on long range observations. The experimental results show that this proposed algorithm can achieve better results than HMM and CRF.