Optimasi Deteksi Pelanggaran Helm pada Traffic Light Berbasis YOLOv7 dan CLAHE

Abstract

Driving safety is a crucial aspect in the transportation system, especially in ensuring the use of helmets by motorcyclists. This research proposes a helmet violation detection system based on the YOLOv7 algorithm optimized with the Contrast Limited Adaptive Histogram Equalization (CLAHE) method to improve performance in low-light conditions. Data is obtained through video recording in traffic light areas and processed through pre-processing stages, such as contrast enhancement, annotation, and augmentation. The training results showed that the YOLOv7 model with CLAHE achieved a precision value of 0.884, recall of 0.873, and mean average precision (mAP) of 0.920 for all object classes. This value is much higher than the YOLOv7 model without CLAHE which only obtained a precision of 0.731, recall of 0.704, and mAP of 0.728. Thus, this approach proves effective in improving the accuracy of helmet violation detection, especially in low-light conditions, and can be applied as a technology-based traffic law enforcement support system.