Detection of Motorcyclists without Helmet from Traffic Video using Deep Learning Techniques
C Drisya, Leena Mary · 2022 IEEE 19th India Council International Conference (INDICON) · 2022
An increase in the vehicle density on the road can cause a rise in the number of accidents, where motorcyclists are more exposed to serious head injuries. Existing detection methods, however, have several disadvantages, including the inefficiency to follow particular motorcycles across multiple frames, and false detection of parked motorcycles in the traffic video. Additionally, variations in traffic density and traffic environments are restricted in the datasets that are used to develop existing approaches. In this paper, we propose a system to automatically detect motorcyclists those who are not wearing helmets from traffic videos using deep learning techniques. Motorcycles are detected and identified using a pre-trained YOLOv4 network and moving objects are detected using background subtraction followed by morphological operations. The detected motorcycles are tracked and top region of bounding boxes are given to a customized CNN classifier to detect motorcyclists with or without helmets. The effectiveness of the proposed approach is illustrated on HELMET dataset recorded in Myanmar.