نوع مقاله : مقاله پژوهشی
نویسندگان
1 کارشناسی ارشد، دانشگاه صنعتی مالک اشتر، تهران، ایران
2 استادیار، دانشگاه علوم و فنون هوایی شهید ستاری، تهران، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Reducing injuries and deaths caused by traffic accidents is always the concern of law enforcement officials and governments. In order to reduce injuries in places where the rate of traffic accidents is high due to motorcycle riders not wearing helmets, significant measures have been taken, including the presence of police officers. All these cases are done by human factors, which may reduce the quality of this monitoring and may not bring the desired result, such as the small number of employees and their fatigue. One of the driving violations is the non-use of helmets by motorcyclists. The solution presented in this research is to use deep learning algorithms to detect the use or non-use of helmets by motorcyclists. In addition to detection, the proposed system can analyze traffic data, including When violations are reduced or increased are used. In this research, three versions 416, 320 and spp of the YOLO_v3 deep neural network have been used to detect helmets, and considering that these networks have already been trained on COCO data, the first 53 layers of the network have been used as transfer learning and 53 layers Next, it was trained based on the data used in this research, and then the performance of these three networks was compared with each other. Finally, the proposed automatic helmet detection system performs the detection operation with 96.08% accuracy after being trained by the Helmet Detection dataset. Also, the model was compared with a number of previous works and more favorable results were obtained.
کلیدواژهها [English]