Embedded Vehicle Detection with YOLOv5n on ESP32-S3 for Urban Traffic Applications

• Danilo J. F. de Souza UEA
• Carlos M. S. Figueiredo UEA

DOI: https://doi.org/10.5753/eniac.2026.29946

Resumo

This paper presents the implementation and evaluation of YOLOv5n models for vehicle detection on urban roads using an ESP32-S3 embedded device. The models were trained at different input resolutions, converted to C code using SynapEdge, and evaluated in terms of offline detection performance and embedded inference time. The model trained and evaluated at 224×224 pixels achieved the best detection performance, with an mAP50 of 0.683, but required the highest inference time on the ESP32-S3. In the evaluated floating-point implementation, the 160×160 configuration represented an intermediate operating point between detection performance and computational cost, with an mAP50 of 0.468 and an average inference time of 13.126 s. The results show that lightweight object detection models can be executed on low-cost embedded devices, although input resolution strongly affects both accuracy and processing time.

Referências

Ashkanani, M., AlAjmi, A., Alhayyan, A., Esmael, Z., AlBedaiwi, M., and Nadeem, M. (2025). A self-adaptive traffic signal system integrating real-time vehicle detection and license plate recognition for enhanced traffic management. Inventions, 10(1):14.

Brink, H., Richards, J. W., and Fetherolf, M. (2016). Real-World Machine Learning. Manning, Shelter Island, NY, 1st edition.

Chang, Y.-H., Wu, F.-C., and Lin, H.-W. (2025). Design and implementation of ESP32-based edge computing for object detection. Sensors, 25(6):1656.

Everingham, M., Gool, L. V., Williams, C. K. I., Winn, J., and Zisserman, A. (2010). The Pascal visual object classes (VOC) challenge. International Journal of Computer Vision, 88(2):303–338.

Leite, C. and Awad, J. C. M. (2012). Cidades Sustentáveis, Cidades Inteligentes: Desenvolvimento Sustentável num Planeta Urbano. Bookman, Porto Alegre, 1st edition.

Loureiro, C. E. N., Guedes, E. B., and Figueiredo, C. M. S. (2025). Monitoramento inteligente de tempo real para tráfego urbano em cidades brasileiras. In Anais do XVII Simpósio Brasileiro de Computação Ubíqua e Pervasiva (SBCUP), pages 41–50, Maceió/AL. SBC.

Moosmann, J., Müller, H., Zimmerman, N., Rutishauser, G., Benini, L., and Magno, M. (2024). Flexible and fully quantized lightweight TinyissimoYOLO for ultra-low-power edge systems. IEEE Access, 12:75093–75107.

Prince, S. J. D. (2012). Computer Vision: Models, Learning, and Inference. Cambridge University Press, Cambridge.

Sapkota, R. and Karkee, M. (2025). Ultralytics YOLO evolution: An overview of YOLO26, YOLO11, YOLOv8 and YOLOv5 object detectors for computer vision and pattern recognition. arXiv preprint arXiv:2510.09653. Disponível em: [link].

Shafi, A. (2025). Running YOLOv5n on an ESP32-S3. Hackster.io. Disponível em: [link]. Acesso em: 29 maio 2026.

Uber Newsroom (2020). Estudo com dados da Uber conclui que congestionamento piora até 32% quando ocorrem acidentes. Uber Newsroom Brasil. Estudo realizado pelo Instituto Cordial. Disponível em: [link]. Acesso em: 26 mar. 2026.

Zhao, L., Song, Y., Zhang, C., Liu, Y., Wang, P., Lin, T., Deng, M., and Li, H. (2020). T-GCN: A temporal graph convolutional network for traffic prediction. IEEE Transactions on Intelligent Transportation Systems, 21(9):3848–3858.