An Edge Efficiency Score for Computer Vision Models Applied to Traffic Sign Detection

• Marco Aurélio Gibellato Nogueira UTFPR
• Rafael Staiger Bressan UFPR
• Teruo Matos Maruyama UTFPR
• Danilo Sipoli Sanches UTFPR
• Alessandro Botelho Bovo UTFPR
• André Robero Ortoncelli UTFPR

DOI: https://doi.org/10.5753/eniac.2026.30859

Resumo

Traffic sign detection is essential for ADAS on resource-constrained edge devices. This work presents a methodology for evaluating traffic sign detectors on TT100K, including YOLO variants, RT-DETR-L, SSD MobileNet, and Faster R-CNN. We propose the Edge Efficiency Score (EES), which combines accuracy, small-target performance, latency, GPU memory, and GFLOPs for deployment-oriented model selection. Results show that YOLOv8s achieved the highest detection performance, while YOLOv8n achieved the highest EES by better balancing accuracy and computational cost. These findings reinforce the need to jointly consider detection quality and operational efficiency in embedded ADAS and ITS applications.

Referências

Chi, X. et al. (2023). Dataset and improved yolov7 for text-based traffic sign detection. In The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, volume XLVIII-1/W2-2023, pages 881–887, Cairo, Egypt.

Gong, C. et al. (2022). Traffic sign recognition based on the yolov3 algorithm. Sensors, 22(23):9345.

Li, M., Zhang, D., He, Q., Zhou, C., Li, M., and Zhou, X. (2026). Msd-yolo: A multi-scale and detail-enhancement network for traffic sign detection. Computers, Materials & Continua, 87(3):1.

Li, Y. et al. (2024). A small-target traffic sign detection algorithm based on partial conv and atrous spatial pyramid. IET Image Processing, 18(15):4639–4652.

Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., and Zitnick, C. L. (2014). Microsoft coco: Common objects in context. In European conference on computer vision, pages 740–755. Springer.

Liu, L., Wang, L., and Ma, Z. (2024). Improved lightweight YOLOv5 based on ShuffleNet and its application on traffic signs detection. PLoS ONE, 19(9):e0310269.

Lopez-Montiel, M. et al. (2021). Evaluation method of deep learning-based embedded systems for traffic sign detection. IEEE Access, 9:101217–101238.

Padaria, A. A. et al. (2023). Traffic sign classification for autonomous vehicles using split and federated learning underlying 5G. IEEE Open Journal of Vehicular Technology, 4:782–793.

Triki, N., Karray, M., and Ksantini, M. (2024). A comprehensive survey and analysis of traffic sign recognition systems with hardware implementation. IEEE Access, 12:129034–129055.

Wang, J. et al. (2023). Vehicle-mounted adaptive traffic sign detector for small-sized signs in multiple working conditions. IEEE Transactions on Intelligent Transportation Systems, 24(11):12242–12254.

Wang, L., Ai, Q., and Jin, H. (2026). Lightweight traffic sign detection algorithm with noise suppression and semantic enhancement. PloS one, 21(1):e0340810.

Yang, J. (2023). Traffic sign detection based on classic visual recognition models. Journal of Physics: Conference Series, 2646(1):012024.