영상 기반의 야외 환경에서 강인한 하이브리드 충돌 경보 시스템

Abstract

Collision Warning System is used widely in Advanced Driver Assistance System to reduce the number of traffic accidents caused by driver inattention. This paper proposed a Collision Warning System that includes moving object detection and collision warning, and is insensitive to shadows and illumination changes. This method uses images from a single camera. This method is a fusion of two detectors: one combines a Focus of Expansion and an optical flow, the other combines a Binary Histogram of Oriented Gradients and AdaBoost algorithm. I call the fusion detection Hybrid detection. The proposed method uses a probability map to merge the two detectors efficiently. Compared to existing systems, the proposed method had a lower false detection and warning rate, and can find the boundary of objects more exactly when shadows or illumination changes exist in images from the camera. The hybrid method had lower computation power than existing method. This Collision Warning System is more suitable than existing methods when images include a variety of surrounding existing error elements such as shadows and illumination changes, cab be used in an outdoor vision system.