Graph-Cut Based Regional Risk Estimation for Traffic Scene

In this study, we investigate the regional risk estimation of drivers for street environment involving different players such as pedestrians, other vehicles, traffic signs, traffic lights, and crosswalks. Various researches focusing on objects regarding traffic have been realized by means of traditional risk estimation. In turn, conventional methods have not presented a realistic solution for drivers at risky regions and moments; whereas, our approach considers emerging risks for a driver due to dynamic actions of street players. A chessboard is devised for representing the street players, each of which carries different potential risks. Every square of the chessboard refers to a partition, which can host one or multiple players. Further, a partition can have different risks for a driver. The proposed model is realized using a graph-cut algorithm for energy minimization. Each partition is considered as a vertex of the graph, which can transfer risks caused by street players. Vertexes are formed via behavior as those of memory cell structures. The memory cells have risk transfer capabilities allowing a driver to determine momentarily risks on urban traffic. Consequently, this captures the regional risk for driver in light of the detected street players as demonstrated through the paper.