열상 이미지 기반 야간 보행자 인식 기술

Abstract

Recently, vehicles are regarded as the most commonly used transportation for daily lives. Safety is very important in driving. To improve safety, this paper proposes a method for night-time pedestrian detection by using thermal video camera, to prevent accidents due to the limitation of human visible range during night time. We apply histogram equalization and morphological erosion operation to find pedestrian candidates. Then we use Histograms of Oriented Gradients (HOG) of the thermal pedestrian image and classify the candidates by using Support Vector Machine (SVM) for detecting pedestrians. Experimental results are promising.