Optimizing Driving Completeness Prediction Models: A Comparative Study of YOLOv7 and Naïve Bayes at Institut Teknologi Sumatera

Abstract

The number of vehicles in Indonesia is increasing every year. The number of motor vehicle accidents in 2022 will be more than 100,000. It is hoped that several regulations regarding motorbike rider equipment will increase awareness of rider safety. By utilizing image recognition technology developed with artificial intelligence, it is possible to create digital image processing models or images that are fast and accurate for detecting driving equipment. The object detection model developed uses a dataset in the form of images of motorists who want to enter ITERA through the main gate. The object detection model will also be integrated with the classification model to create a program that can detect motorbike rider equipment, such as mirrors, helmets, not wearing a helmet, shoes, not wearing shoes, open clothes, and closed clothes. After detecting motorized rider equipment in the classification area, the results will be transferred to a classification model to determine the level of safety for motorized riders, either insufficient or sufficient safety. The test results show that the optimal object detection model was found at an epoch value of 70 with a batch-size of 16, producing a mAP value of 0.8914. The optimal classification model uses the naive Bayes method which has been trained with a dataset of 62 data and achieves an accuracy of 94%.

Keywords

• model
• yolov7
• naive bayes
• object detection
• driving equipment