A system for determining whether kickboard users wear helmets using YOLOv4
The rapid spread of electric kickboards has led to a surge in accidents caused by electric kickboards in two years. Accordingly, a mandatory helmet was recently enacted when operating an electric kickboard. However, there are still many people who do not wear helmets, and it is difficult to crack down on them only with manpower. In this situation, the solution will be the development of an artificial intelligence helmet monitoring program using a camera. To this end, we will use YOLOv4 to determine whether kickboard users use helmets and save time when determined not to wear them to design a system that gives penalties to users through collaboration with shared kickboard companies.
According to a survey, 7 out of 10 users of personal mobile devices (PMs) such as electric kickboards do not wear helmets. On the 5th, the Korea Transportation Safety Authority announced that "the actual compliance rate of users is still low compared to the awareness of laws related to personal mobile devices (PM) and driving conditions revised in May." According to the 2021 Traffic Culture Index Pilot Survey, 89.8% of users were required to wear safety helmets, but only 26.3% of users actually wore safety helmets. In particular, the compliance rate for wearing safety helmets also differed significantly depending on the type of personal mobile device ownership. In the case of individual owners, the compliance rate for wearing safety helmets was 55.6%, but the compliance rate for shared electric kickboards (mobile devices) was only 13.2%. As can be seen from the above article, the proportion of electric kickboard users without helmets is quite high. There are quite a few users who do not wear safety helmets even though they are aware of the mandatory wearing of safety helmets. However, it is not an easy situation to crack down on. There is a limit to the need for too much manpower for an effective crackdown to take place.
Real-time images are used to identify people who are not wearing helmets, record time, and use them as part of a system that collaborates with a shared kickboard company to penalize those who are not wearing helmets
1. Image collection and labeling
Images of electric kickboard users wearing helmets and electric kickboard users wearing helmets are captured on YouTube videos or collected through Google search. Then, using the labeling program, the objects in each image are divided into with_helmet and without_helmet for boxing. In this way, a total of 193 images were labeled.
2. Imgaug
There is a limit to collecting images manually. If the number of images is not reached at a certain level, the recognition rate is significantly lowered. To this end, a process of increasing the number of images by modifying existing images is required. We changed the number of images using the Imgaug of github. By changing the angle of the picture or the size of the blur effect picture, a total of 3,860 images and labels were obtained, up 20 times from the previous one.