混合式機器學習方法於施工人員個人安全裝備即時辨識之應用

Application of the Hybrid Machine Learning Techniques for Real-time Identification of Worker’s Personal Safety Protection Equipment

營建施工意外一直高居世界各國產業職災之首，究其原因在於營建工地具 有高度開放與動態特性，並受限於工地職安人員之質與量，常難以落實風險控 管與即時防治。然而，受惠於人工智慧 （AI） 深度學習技術快速發展，在工地 物件之動態視覺辨識功能上得到重大突破，使得營建工地職安管理出現新的發 展契機。本研究旨在以人工智慧視覺辨識技術為基礎，發展營建工地施工人員 定位與安全裝備辨識之功能，自動發掘工地勞工之潛在安全危害，並降低職安 管理人員之工作負擔。本研究建構完成之系統適用於營造現場之勞工安全裝備 辨識，並以實際公共工程專案工地進行系統驗證，系統最終訓練結果為召回率 95%以上，精確率 93%以上，實測之正確率為 90%以上，純淨度為 80%以上， 已達輔助職安人員進行安全管理之實用性，可有效提升營建工地安全效益及降 低職災發生之風險。

Construction accidents are the most significant contributor to occupational disasters among all industries worldwide. This is due to both the open and dynamic characteristics of construction sites as well as the insufficient quantity and quality of site safety management personnel. The advancement of Artificial Intelligence (AI) deep learning techniques in dynamically identifying the moving objects on-site offers a promising opportunity to improve construction safety. This paper presents the application of the most state-of-the-art AI techniques to identify on-site construction safety hazards in order to prevent risk events for construction workers. The proposed method has been implemented in a real construction project and achieved satisfactory performance with 95% of Recall, 93% of Precision for lab testing, 90% of Correctness and 80% of Cleanness for insitu testing. It has been concluded that the proposed method has promising potential to assist construction safety management personnel in improving the safety management practices.