A Method for Calculating Future Behaviors of Vehicles toward Effective Collision Prediction

There are many accidents in traffic intersections because of complicated situations. In order to prevent the accidents, collisions between vehicles should be predicted and drivers should be alerted by in-car systems as soon as possible. Our research viewpoint is to detect possible collisions by predicting vehicle behaviors. We propose a region-based approach that uses “attainable region” for predicting the behaviors. In our method, traffic situations are observed by analyzing video streams of the scenes. Vehicle behaviors are transformed into the scene plane as spatial regions. With attainable regions, possible collisions between vehicles can be estimated by checking overlaps between regions. Through evaluation experiments, we show the feasibility of prediction with attainable regions.