Detecting Helmets on Motorcyclists by Deep Neural Networks with a Dual-Detection Scheme

This paper deals with the problem of helmet detection on motorcyclists in single images. Some previous attempts have been made to detect helmets on motorcyclists, most of which are designed for videos and not suitable for single-image helmet detection. In this paper, we propose a single-image motorcyclist helmet detection method via deep neural networks with a dual-detection scheme. Two types of detectors are first trained, namely, the rider detection and the head-shoulder detection. A post-processing strategy is devised to remove duplicate detections and pedestrians (without motorbikes). Furthermore, we present a new benchmark dataset named SCAU helmet detection on motorcyclists (SCAU-HDM). Extensive experimental results demonstrate the effectiveness of the proposed method.