Keypoint-based binocular distance measurement for pedestrian detection system
Abdelhamid Mammeri, Azzedine Boukerche, Mingchang Zhao · 2014
Pedestrian Detection System (PDS) has become a significant research area designed to protect road-users. Despite the huge number of research works, many PDSs are designed to detect pedestrians without knowing their precise distances from vehicles. In fact, a priori knowledge of the distance between the car and pedestrian allows taking the appropriate decision to avoid collisions. In this context, the distance estimation problem is investigated in this paper. For that purpose, we use HOG-SVM method to detect pedestrians, and keypoints-based feature extraction method to generate the parallax in a binocular vision system. In addition, we design a crossover re-detection method to reinforce the robustness of the system. Our approach is not only applicable to measure the distance between vehicles and pedestrians, but, it can be used efficiently to measure the distance to other objects such as traffic signs or animals. Through real word experiments, the system shows a margin of error of 7.5 % in outdoor and long-range conditions.