Region-Based Trajectory Analysis for Abnormal Behaviour Detection: A Trial Study for Suicide Detection and Prevention

Xun Li, Ryan Anthony J. de Belen, Arcot Sowmya, Sandersan Onie, Mark E. Larsen

Abstract: We propose a region-based trajectory analysis method to detect abnormal activities in a scene. It provides a self-adapted, location-sensitive and interpretable trajectory analysis method for different scenarios. Our integrated pipeline consists of a pedestrian detection and tracking module to extract density, speed and direction features. In addition, it contains a grid-based feature extraction and clustering module that automatically generates a region map with corresponding feature importance. During testing, the pipeline analyses the segments that fall into the regions, but do not comply with the important features, and clusters trajectories together to detect abnormal behaviours. Our case study of a suicide hotspot proves the effectiveness of such an approach.