Unsupervised detection of abnormal behavior for intelligent surveillance

作者：Zhang Changqi; Guan Yepeng^*

来源：Journal of Information and Computational Science, 2014, 11(13): 4805-4818.

DOI：10.12733/jics20104372

• Abnormal behavior
• Abnormal behavior detections
• Dynamic flow field
• Intelligent surveillance
• Motion features
• Particle flow
• Unsupervised detection

A novel method based on dynamic particle flow has been developed to detect abnormal behavior automatically. The proposed approach starts by taking motion crowds as an aperiodic dynamical flow field to solve abnormal behavior detection problem. The motion in the scene is described by the particle trajectories computed by time integration of the dynamical flow field. Abnormal behavior is detected automatically based on some high-level features extracted from the developed particle flow without any hypothesis for the scene conditions in advance. The proposal does not require any object detection, tracking, or training in detecting abnormal behaviors. Comparative studies with some investigated methods have indicated the superior performance of the proposal in detecting abnormal behaviors from cluttered contents with dynamic scene variation and crowded environments.

• 出版日期2014
• 单位上海大学