<![CDATA[Low lighting conditions in CCTV images cause a decrease in the accuracy of the Electronic Traffic Law Enforcement (E-TLE) system, especially in detecting helmet use among motorcyclists. Dark images with low contrast and high noise hinder the feature extraction process, so that deep learning-based detection models often produce False Negatives. This study proposes the integration of the Contrast Limited Adaptive Histogram Equalization (CLAHE) preprocessing method with the YOLOv8 architecture to improve the performance of helmet violation detection in low-light environments. The Helmet Detection dataset is used with the addition of synthetic low-light augmentation to simulate variations in nighttime lighting intensity. Tests show that the use of CLAHE can significantly improve the quality of visual features, as evidenced by the increase in Mean Average Precision (mAP@0.5) from 72.4% in raw images to 89.1% after preprocessing. In addition, the system is still able to operate in real-time with an average speed of 35–37 FPS on a Tesla T4 GPU. These results indicate that the integration of CLAHE and YOLOv8 is effective in improving the reliability of helmet violation detection in low-light conditions and is feasible to be implemented in computer vision-based traffic surveillance systems.]]>
Copyrights © 2026
