營建產業因作業環境複雜且危險性高，長期以來職業災害發生率居高不下，其中墜落與未配戴個人防護具為造成重大傷亡之主因。然而，現有主流公開資料集多為光線充足、背景單純之靜態影像，難以反映台灣道路施工常面臨的夜間低照度、強光眩光、動態遮擋及雨天等真實挑戰，導致模型在實務應用上會產生嚴重的場域落差。
針對上述問題，本研究目的在建立一套適用於台灣在地化營建場域的電腦視覺偵測方法。研究採用 YOLOv8 模型，針對自行蒐集之台北市道路挖掘管理中心公開施工影像進行訓練與評估。為確保資料品質，本研究建立了一套完整的資料工程流程，包含去除高相似度影格，並制定嚴謹的標註策略以過濾無效樣本。資料集涵蓋日間 (Day)、夜間 (Night)、夜間紅外線 (IR) 及雨天 (Rainy) 等多樣化情境。
實驗結果顯示，本研究提出的方法在複雜場景下展現了普遍的適應性。特別是在缺乏色彩資訊的夜間紅外線 (IR) 場景中，模型之平均精確度 (mAP@0.5) 達到 0.72，與日間場景的 0.71 表現相當。 此結果證實紅外線影像能有效克服夜間光源不足的問題，顯著提升人員輪廓的辨識度。儘管研究發現夜間低光與強光眩光及遮擋、小物件仍是造成漏檢的主要因素，但本研究建立的在地化資料集與分析流程，已驗證YOLOv8 應用於全時段工地監控的可行性，並為未來的智慧營造技術提供了具體的實務參考。

This study develops and evaluates a safety-helmet wearing recognition approach for realworld construction-site monitoring in Taiwan under complex illumination and dynamic scenarios.Unlike common public datasets that mainly contain well-lit, static images with clean backgrounds, road construction videos often suffer from low light, glare from artificial lighting and vehicle headlights,rain-induced blur and reflections, motion blur, and frequent occlusion among workers and machinery.These factors create a domain gap that limits direct deployment of vision models trained on standard benchmarks. To address this gap,we construct a localized dataset from 528 publicly available roadwork videos (217.23 hours) and train a YOLOv8 detector with a strict data engineering and annotation protocol.Two classes are defined: Person_with_helmet and Person_without_helmet, using full-body person bounding boxes to better preserve contextual cues when head regions are small or ambiguous.The model achieves stable 5-fold cross-validation performance (mAP@0.5 = 0.777 ± 0.063). On the held-out test set at confidence threshold 0.25, it reaches Precision = 0.72, Recall = 0.68, and mAP@0.5 = 0.68.Scenario-based evaluation further reveals that visible-light nighttime is the most challenging condition, while IR nighttime can provide comparable performance to daytime.Rainy scenes mainly reduce recall due to missed detections. These results demonstrate the feasibility of applying YOLOv8 to all-day construction monitoring and provide practical insights for future improvements in robust PPE compliance inspection.

[1] 勞動部職業安全衛生署, "112年勞動檢查統計年報," 勞動部職業安全衛生署,臺灣, 2023. Accessed: 2025-07-10. [Online]. Available:https://www.osha.gov.tw/48110/48331/48333/48339/173234/post
[2] R. Girshick, J. Donahue, T. Darrell, and J. Malik, "Rich feature hierarchies for accurate object detection and semantic segmentation," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2014, pp. 580–587.
[3] S. Ren, K. He, R. Girshick, and J. Sun, "Faster r-cnn: Towards real-time object detection with region proposal networks," Advances in neural information processing systems, vol. 28, 2015.
[4] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You only look once: Unified, real-time object detection," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 779–788.
[5] G. Jocher, A. Chaurasia, and J. Qiu, "Ultralytics yolo (version 8.0. 0)[computer software]," ed, 2023.
[6] J. Wu, N. Cai, W. Chen, H. Wang, and G. Wang, "Automatic detection of hardhats worn by construction personnel: A deep learning approach and benchmark dataset," Automation in construction, vol. 106, p. 102894, 2019.
[7] N. D. Nath, A. H. Behzadan, and S. G. Paal, "Deep learning for site safety: Realtime detection of personal protective equipment," Automation in construction, vol. 112, p. 103085, 2020.
[8] Z. Wang, Y. Wu, L. Yang, A. Thirunavukarasu, C. Evison, and Y. Zhao, "Fast personal protective equipment detection for real construction sites using deep learning approaches," Sensors, vol. 21, no. 10, p. 3478, 2021.
[9] K. Kim, K. Kim, and S. Jeong, "Application of YOLO v5 and v8 for recognition of safety risk factors at construction sites," Sustainability, vol. 15, no. 20, p. 15179, 2023.
[10] M. Ferdous and S. M. M. Ahsan, "PPE detector: a YOLO-based architecture to detect personal protective equipment (PPE) for construction sites," PeerJ Computer Science, vol. 8, p. e999, 2022.
[11] T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollár, and C. L. Zitnick, "Microsoft coco: Common objects in context," in European conference on computer vision, 2014: Springer, pp. 740–755.
[12] T.-Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, "Feature pyramid networks for object detection," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 2117–2125.
[13] W. Hua and Q. Chen, "A survey of small object detection based on deep learning in aerial images," Artificial Intelligence Review, vol. 58, no. 6, pp. 1–67, 2025.
[14] S. Zhang, R. Benenson, and B. Schiele, “CityPersons: A Diverse Dataset for Pedestrian Detection,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2017, pp. 4457–4465.
[15] Z. Ouardirhi, S. A. Mahmoudi, and M. Zbakh, "Enhancing object detection in smart video surveillance: A survey of occlusion-handling approaches," Electronics, vol.13, no. 3, p. 541, 2024.
[16] K. Saleh, S. Szénási, and Z. Vámossy, "Occlusion handling in generic object detection: A review," in 2021 IEEE 19th World Symposium on Applied Machine Intelligence and Informatics (SAMI), 2021: IEEE, pp. 000477–000484.
[17] Y. P. Loh and C. S. Chan, "Getting to know low-light images with the exclusively dark dataset," Computer vision and image understanding, vol. 178, pp. 30–42, 2019.
[18] C. Shorten and T. M. Khoshgoftaar, "A survey on image data augmentation for deep learning," Journal of big data, vol. 6, no. 1, pp. 1–48, 2019.
[19] Y. Wang, W. Xie, and H. Liu, "Low-light image enhancement based on deep learning: a survey," Optical Engineering, vol. 61, no. 4, pp. 040901–040901, 2022.
[20] Z. Chen, Z. Zhang, Q. Su, K. Yang, Y. Wu, L. He, and X. Tang, "Object detection for autonomous vehicles under adverse weather conditions," Expert Systems with Applications, p. 128994, 2025.
[21] C. Sakaridis, D. Dai, and L. Van Gool, "ACDC: The adverse conditions dataset with correspondences for semantic driving scene understanding," in Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 10765–10775.
[22] H. Song, X. Zhang, J. Song, and J. Zhao, "Detection and tracking of safety helmet based on DeepSort and YOLOv5," Multimedia Tools and Applications, vol. 82, no.7, pp. 10781–10794, 2023.