With the rapid expansion of smart cities and growing demands for traffic safety and public surveillance, the importance of CCTV-based object detection technologies has significantly increased. This study proposes an enhanced object detection framework based on the latest YOLOv11 model, optimized for real-world traffic environments. Key strategies include: (1) maintaining the original aspect ratio of input images to reduce object distortion, (2) applying data augmentation techniques simulating adverse conditions such as fog and motion blur, and (3) conducting comparative experiments across different YOLOv11 model configurations. Using a dataset of 7, 000 annotated CCTV images collected from the Dongan intersection in Pangyo Zero City, the YOLOv11s-Fog model achieved outstanding performance, with an mAP50 of 0. 97 and precision of 0. 97. Notably, the model showed significant improvement in detecting small-scale objects such as pedestrians (mAP50ₕuman = 0. 87), effectively addressing limitations found in previous YOLO models. These results demonstrate the model's robustness in visually challenging conditions and suggest its high applicability in pedestrian-prioritized areas such as school zones and senior protection zones. The proposed approach contributes to improving the reliability of CCTV-based monitoring systems and enhancing real-time risk detection in smart city environments.
Jeon et al. (Thu,) studied this question.