Abstract
Driving hazardous factors or dangerous regions detection algorithm is proposed This algorithm is based on two combinations of local information, provided by optical flow, and global information, obtained from vehicle's states. The algorithm infers degrees of danger of moving objects in dynamic images taken from a vehicle-mounted camera using artificial neural networks. It makes the system more flexible to multiple drivers with dissimilar characteristics. Experiments are done on model and real vehicles videos with multiple moving objects. The results show that it is possible to determine dangerous regions close to the drivers' decisions. The foundation of a practical driver assistance system is fixed by the proposed algorithm.
