A Comparative Study of YOLOv4-Nano (Darknet) and Ultralytics YOLOv7, YOLOv8, YOLOv10, and YOLOv26 for Real-Time Industrial Vision: PPE Detection, Human Safety Monitoring, and Surface Defect Inspection

This paper has presented a systematic comparative study of YOLOv4-Nano (Darknet framework) and YOLOv7, YOLOv8, YOLOv10, and YOLOv26 (Ultralytics framework) for three simultaneous real-time industrial vision applications: PPE compliance detection, human safety zone monitoring, and manufactured surface defect inspection. Deployed on a PC-class GPU server with direct PLC integration and local SQL event logging, the system demonstrates sustained real-time throughput exceeding 30 FPS across all three concurrent tasks, with PLC actuation latency below 100ms.   The central finding is that task-optimised multi-model deployment — selecting YOLOv8-m for PPE, YOLOv7 for human monitoring, and YOLOv26 for defect detection — consistently outperforms any single-model deployment across all three tasks. The Ultralytics framework delivers substantially lower integration engineering effort than Darknet for production environments requiring PLC connectivity and SQL logging, while achieving comparable inference throughput via TensorRT FP16 optimisation. YOLOv26's Transformer-enhanced backbone delivers the most significant capability improvement for visually subtle defect classes, representing the recommended upgrade path for defect inspection systems currently running earlier YOLO versions.   The results and architecture presented in this paper are directly applicable to industrial engineers deploying YOLO-based vision systems across manufacturing safety, quality inspection, and Zone monitoring use cases, providing evidence-based model selection guidance grounded in production measurements rather than benchmark-only evaluation.

[1] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You Only Look Once: Unified, Real-Time Object Detection," in Proceedings of CVPR, 2016, pp. 779–788. [2] J. Redmon and A. Farhadi, "YOLOv3: An Incremental Improvement," arXiv:1804.02767, 2018. [3] A. Bochkovskiy, C.-Y. Wang, and H.-Y. M. Liao, "YOLOv4: Optimal Speed and Accuracy of Object Detection," arXiv:2004.10934, 2020. [4] C.-Y. Wang, A. Bochkovskiy, and H.-Y. M. Liao, "YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real-Time Object Detectors," in Proceedings of CVPR, 2023. [5] G. Jocher, A. Chaurasia, and J. Qiu, "Ultralytics YOLOv8," GitHub Repository, 2023. [Online]. Available: https://github.com/ultralytics/ultralytics [6] A. Wang, H. Chen, L. Liu, K. Chen, Z. Lin, J. Han, and G. Ding, "YOLOv10: Real-Time End-to-End Object Detection," arXiv:2405.14458, 2024. [7] Y. Nath and K. Rohit, "Deep Learning Based PPE Detection in Construction and Manufacturing Sites: A Systematic Review," Safety Science, vol. 155, 2022. [8] S. Fang, X. Li, and Q. Hu, "Real-Time Human Detection and Zone Monitoring in Industrial Environments Using Deep Learning," IEEE Transactions on Industrial Informatics, vol. 18, no. 4, 2022. [9] X. Zhao, T. Zhang, and B. Du, "Deep Learning-Based Surface Defect Detection: A Survey," IEEE Transactions on Instrumentation and Measurement, vol. 70, 2021. [10] L. Chen and R. Kumar, "Vision-PLC Integration for Smart Manufacturing: Architecture and Latency Analysis," Journal of Manufacturing Systems, vol. 62, pp. 110–122, 2022.