<![CDATA[License plate detection is a crucial component of intelligent transportation systems. Deep learning methods still face limitations in detecting small-sized plates under low-light conditions and complex backgrounds. This study evaluates YOLOv12's performance for license plate detection in CCTV imagery containing small objects with great visual detail. Unlike YOLOv11, which focuses on detection efficiency for larger objects, YOLOv12 integrates attention mechanisms to enhance sensitivity to fine-grained spatial features. Model evaluation was conducted using precision, recall, and mean average precision (mAP) metrics on traffic image datasets with daytime and nighttime lighting conditions and CCTV viewing angles. Results show the model achieves mAP@0.5 of 87.2% and precision of 89.5%, comparable to previous YOLO-based studies. However, performance drops to 47.9% at mAP@0.5:0.95, indicating limitations in bounding-box localization precision under visually complex conditions. This study highlights opportunities for future improvement through dataset expansion and parameter optimization for training.]]>
Copyrights © 2026
