Severity detection of traffic accidents at intersections based on vehicle motion analysis and multiphase linear regression

This paper proposes a new approach to describe traffic scene including vehicle collisions and vehicle anomalies at intersections by video processing and motion statistic techniques. The research mainly targets on extracting abnormal event characteristics at intersections and learning normal traffic flow by trajectory clustering techniques. Detecting and analyzing accident events are done by observing partial vehicle trajectories and motion characteristics. The model implements video preprocessing, vehicle detection and tracking in order to extract motion characteristics through vehicle existence on road lanes. Activity patterns are determined by trajectory clustering analysis. Normal and abnormal traffic events are segregated by using log-likelihood thresholds. Abnormal traffic events and collisions are characterized using linear multiphase regression analysis technique, which apply semantic information extraction about traffic incidents.