營建勞工職業安全災害之發生,在世界各國一直是無法完全杜絕之問題。尤其在追求施工進度過程中,營建勞工安全經常被忽略,而重大意外事故中以墜落及滾落為最大宗,而墜落發生之地點又以開口為最。為降低勞安事故發生,本研究提出開口防護設施安全辨識系統基於深度學習技術,可輔助施工現場環境安全的風險監測,監測具有風險的開口,結合監視攝影設備及一張低成本貼紙,即可實現24小時監控的功能。研究採用更快速區域卷積神經(Faster R-CNN)網路技術辨識開口的位置及其安全狀態,在對其偵測的結果進行安全狀態的分類。在實驗室測試中,召回率達100%,精確率達95.36%;實地案例測試中召回率亦超過95%,精確率也達87.91%。本研究所提出之基於影像辨識的開口防護設施安全辨識方法,能夠在營建複雜施工環境中輔助監視開口防護設施的安全狀態,以達輔助檢查人員的效果,不但能減輕檢查人員的負擔,更能有效快速的通報相關人員進行安全處置。
Construction accidents are inevitable not only in Taiwan but also in many other countries. Unfortunately, the construction safety is usually overlooked, especially when the schedule is in hurry. Nevertheless, the construction industry has contributed the major fatal occupational accidents among other industries. Falls in unprotected oppresses have long been the primary contributors to severe construction accidents. As a result, this research aims at developing a method for automated identification of unprotected construction openness on site based on a Faster-RCNN technique. By combining an affordable site camera and a specially designed tag, the proposed method can achieve all-time monitoring of the safety status of construction opening on site with the proposed Faster-RCNN technique. The experimental results show that the proposed method achieves 100% Recall and 95% Precision in identifying unprotected construction opening in lab; while it achieves 95% Recall and 87.91% Precision in real-world construction site testing. It is concluded that the proposed method has profound potential for practical implementation. It is also promising to assist the construction safety personnel in identifying the unsafe site conditions promptly, as such more construction accidents can be prevented in time.