人工智慧視覺化技術於工地危害辨識之應用

Application of AI-based Computer Visualization Techniques to the Hazard Identification of Construction Site

本研究旨在以最新發展之人工智慧視覺辨識技術為基礎，發展營建工地安全危害辨識之功能，開發智慧型勞工安全視覺辨識系統，自動發掘工地勞工之潛在安全危害。期能在安全事故發生之前，透過事先之風險辨識與處置，以改善目前居高不下的工地勞安意外問題，並降低職安管理人員之工作負擔。本研究完成項目包括危害辨識技術及新科技運用現況之回顧、工地即時影像串流之擷取技術開發、分析並設計適用於營造現場之智慧型勞工安全視覺辨識雛型系統（CISDS）等工作，並以實際公共工程專案工地進行系統驗證及危害辨識結果評估。經過實際專案測試結果發現，本研究所提出之方法在系統訓練階段之召回率皆超過95％，而精確率皆可達到93％以上；現場實測純淨度超過90％，而正確率約可達80％以上。本研究所提出之智慧型勞工安全視覺辨識雛型系統，可進行安全裝備辨識、電梯直井開口墜落預防及結構開口辨識等功能，能提升營建工地安全管理之效益，具有產業推廣之潛力，未來可以推廣至各公私部門建築工程之業主、監造及施工相關單位之職安管理人員應用。

Construction accidents have been the main contributor of the vocational disasters for all industries worldwide. It's due both to the open/dynamic characteristics of the site and to the limitation of quantity and quality of site safety managers. Thanks to the advancement of Artificial Intelligence (AI) based visualization techniques in dynamically identifying the moving objects on-site, the construction safety is apt to be improved nowadays. The current project aims at applying the most state-of-the-art AI techniques to identify the construction safety hazards on site, so that the risk events can be prevented. The project has reached the following results: the summarization of advanced hazard identification technologies, the development of techniques for image capturing from video streams, the programming of an AI-based Intelligent Safety Detection System for Construction (CISDS). The developed CISDS was verified and evaluated via real-world case projects to test its feasibility and reliability. The lab testing of all modules revealed that 95% of Recall and 93% of Precision were achieved; the in-situ testing reported that around 90% of Cleanness and 80% of Correctness were achieved. In summary, a prototype of AI-based computer visualization system (namely CISDS) to the hazard identification of construction site has successfully developed, it has been tested to provided profound potential in improvement of the construction safety management. Future users of the proposed system may include the construction safety managers of the project owners, supervision consultants, and general contractors both for public and private building construction projects.