Muh Fajrin Bakri, Shahnaz Tasha Kurnia, Muhammad B. Fajar, Andi Baso Kaswar, Dyah Darma Andayani, Fhatiah Adiba, Sanatang, Syahrul
This study aims to improve the accuracy of legal and illegal motorcycle parking classification by proposing a computer vision-based detection system using YOLOv12 segmentation and region-of-interest (ROI)-guided orientation analysis. The proposed system integrates object segmentation, ROI mapping, and computation of the angular deviation between vehicle orientation and parking guides. To determine the optimal orientation tolerance, experiments are conducted using multiple angular thresholds under identical datasets and testing scenarios. The system is evaluated across three locations using confusion-matrix-based performance metrics. The results show that a 30° tolerance yields the best performance, achieving an average accuracy of 87.29%, precision of 93.14%, recall of 92.23%, and F1-score of 92.68%. These findings indicate that ROI-guided orientation analysis enhances the reliability of motorcycle parking violation detection. Copyright© by the authors.
Department of Computer Engineering, State University of Makassar, Makassar, Indonesia
