Real time detection of stopped vehicles in traffic scenes

Abstract

Computer vision techniques are widely employed in Traffic Monitoring Systems (TMS) to automatically derive statistical information on traffic flow and trigger alarms on significant events. Research in this field embraces a wide range of methods developed to recognize moving objects and to infer their behavior. Tracking systems are used to reconstruct trajectories of moving objects detected often by using background difference approaches. Errors in either motion detection or tracking can perturb the position of the object centroids used to build the trajectories. To cope with the unavoidable errors, we have conceived a method to detect centers of non-motion through recognizing short stability intervals. These are further connected to build the long stability interval used to measure the overall vehicle stopping time. Extensive experiments also accomplished on the sequences provided by AVSS 2007 prove the effectiveness of our approach to measure the maximum stopped delay, even through a comparison with the ground truth.