부산대학교 도서관

This paper proposes a forward vehicle (cars and motorcycles) deceleration detection system for motorcycles at nighttime. Compared with cars, motorcycles are more unstable and dangerous for drivers. However, vision-based driver assistance systems are seldom developed for motorcycles because the tilt rotation of the cameras on the motorcycles cause difficulties. Our proposed method solves this problem to stably detect the deceleration of forward vehicles at nighttime. The proposed system has four main stages: region of interest (ROI) adjustment and idling detection, taillight detection, taillight tracking, and brake light detection. The system first dynamically adjusts the size and position of the ROI and then detects the motions of forward vehicles using the Lucas-Kanade optical flow method. Second, the system detects the candidates of vehicle taillights and uses a support vector machine (SVM) model to identify and locate the position of taillights. Third, a Kalman filter technique is used to track the detected taillights. Since the taillights and brake lights of many vehicles are designed to overlap or be or very close, this step helps to detect the brake lights. Finally, a brake-light-activated threshold adjustment algorithm is applied to detect the starting moment of the brake lights of forward vehicles. The experimental results show that this proposed method obtains a stable result for forward vehicle detection for the motorcycle at nighttime.