Prediction of collisions between vehicles using attainable region
Noritaka Sekiyama, Kazuma Minoura, Toyohide Watanabe · 2011
Traffic surveillance is an important facet of computer vision. In this paper, we propose a method for predicting collisions between vehicles from traffic monitoring movies. In our method, future behaviors of vehicles are transformed into spatial regions called Attainable Regions (ARs). Collisions are predicted by checking whether these two different ARs overlap each other. From experimental results, we show the feasibility of developing a collision-warning system by using ARs.