Detector ensemble based on false positive mining for pedestrian detection

In recent years, the demands for Advanced Driving Assistance Systems (ADAS) is increasing, and pedestrian detection methods using an in-vehicle camera have been widely studied. In the case of pedestrian detection using an in-vehicle camera, since road environment varies widely, it is very difficult to do so accurately by a single classifier. This wide variety could also be understood as large intra-class variation of backgrounds, which leads to the increase of over-detections. To overcome this problem, this paper proposes a method of pedestrian detection using environment clustering based on false detection tendencies. By analyzing the false detection tendency in each environment, the proposed method creates classifiers that can cope with false detections observed in the specific environment. In addition, detector ensemble is introduced to extend this idea for handling multiple environments at the same time. To evaluate the effectiveness of the proposed method, experiments were conducted on the Daimler mono benchmark datasets. Results showed that the proposed method outperformed the conventional methods.