DETECTION SYSTEM OF MOTORCYCLE USER VIOLATIONS WITHOUT HELMETS USING THE YOLO ARCHITECTURE CNN METHOD BASED ON EMBEDDED SYSTEMS

Abstract

Safety Driving is a critical aspect of daily life that cannot be ignored. Riding a motorized vehicle, such as a motorcycle or car, carries a number of risks, and it is important to adopt appropriate safety measures. The use of helmets is one of the measures to improve safety in driving, this is regulated in Law number 22 of 2009 concerning Road Traffic and Transportation (LLAJ). However, in everyday life there are still many motorcycle users who violate these regulations. Security officers often find it difficult to identify violations that occur due to the disproportionate number of violators and security officers. The application of Machine Learning CNN method YOLO architecture is believed to be able to help security officers in identifying violations that occur. This program is made in the form of an embedded system in the form of a Raspberry Pi 4. From the results of training the model using 1056 images with 30 epochs , the results of the accuracy of the model itself is at 74.9%. Testing the system itself using the Blackbox method of 5 features tested shows valid results, but the FPS measurement only gets an average of 1.76 frames per second when the system is run. This shows that the system has met the functionality but has not met the performance.