Stereo-based region of interest generation for real-time pedestrian detection

Abstract

Pedestrian detection is one of the major goals in advanced driver assistance systems (ADAS) which has become an active research area in recent years. In this paper, we present a stereo based pedestrian detection system by fusing the depth and color data provided by a stereo vision camera on a moving platform. The proposed method uses an adaptive window for region of interest (ROI) generation using dense depth map. The extracted candidates are then applied to a Histogram of Oriented Gradients (HOG) feature descriptor to refine ROIs and Support Vector Machine (SVM) is used to classify them into pedestrian and non-pedestrian classes. The system is tested on a stereo based DAS dataset and results show that our system is able to detect pedestrians with different scales and illumination conditions and in presence of partial occlusion.