KAZE-BOF-based large vehicles detection at night

Abstract

Instead of a high-priced radar system, this paper proposes a camera-based system to detect large vehicles at night. Recognition rate is improved using KAZE-based Bag-of-Features component in the system. In our current setup, ROI are extracted from nighttime-driving videos using rear-lights detection, then a LBP-based Adaboost detector is used to detect the locations of large vehicles, finally the KAZE-based BOF classifier is used to improve classification accuracy. Preliminary experimental results shows that this idea is feasible and can help enhance the driver's safety while driving at night.