Safety quantification of intersections using computer vision techniques

Abstract

Vision-based safety analysis is a difficult task since traditional motion-based techniques work poorly when pedestrians and vehicles stop due to traffic signals. This work presents a tracking method in order to provide a robust tracking of pedestrians and vehicles, and quantify safety through investigating the tracks. Surrogate safety measurements are estimated including TTC and DTI values for a highly cluttered video of Las Vegas intersection and the performance of the tracking system is evaluated at detection and tracking steps separately.