職場危害智慧影像辨識技術應用──以綜合商品零售業為例

Regarding Violence in Workplace Hazard Intelligence Image Recognition Technology Application - A Case Study in the Comprehensive Retail Industry

有鑑於近期發生多起超商暴力事件，零售業勞工作業安全引起重視，為提升零售業作業人員發生緊急事件時之報案時效，勞動部勞動及職業安全衛生研究所（簡稱勞安所）整合影像處理辨識技術，建立影像辨識危害鑑別運算模組，整合顧客入內口罩自動辨識、櫃檯作業人員異常姿態偵測、管制區域人員入侵偵測，以及研發超商店員制服辨識功能。透過智慧影像辨識技術，當異常事件發生時（如：店員異常姿態、管制區域遭入侵），提供自動示警，並建立報案機制。
在影像辨識模型訓練部分，採用情境模擬方式蒐集作業勞工可能的各種姿態，以開發前述四種辨識功能。為實行警政單位自動通報，本研究建立警政自動通報機制，當異常事件發生時，程式自動合成報案語音檔，進行語音報案，並自動撥打語音報案電話至超商所屬之分局專線，警方接獲報案後依報案內容進行聯繫確認，以完成報案程序。另外，當異常事件發生時，系統會事先將異常事件發生當下畫面、時間、地點及事件樣態等資訊傳送給超商專責管理人員進行二次確認，以提高異常事件通報的準確性。為測試影像辨識及自
動報案機制之成效，於3家零售業作業場所進行實際場域技術測試，與商家溝通協調監視攝影機架設位置及完成網路設定，完成實地場域的影像辨識危害鑑別運算模組功能驗證，
並依據超商所在地，完成警政通報機制之測試，由分局員警回撥電話完成通報驗證。
有鑑於影像辨識及智慧科技之應用已日趨成熟，零售業可使用此技術作為職場危害監測工具之一，未來業者可依實際需求調整及延伸更多辨識項目，以達到職場危害預防之目的。本研究以透過智慧科技即時偵測零售業作業場所之異常行為為例，建立自動報案機制，強化與警方間立即緊急通報模式，讓警方可更即時到現場處理案件，減少通報所需時間以提高報案效率。

Abstract
In light of the recent surge in violent incidents at convenience stores, there is a heightened concern for the occupational safety of retail industry workers. To improve the efficiency of reporting emergencies in the retail industry, the Ministry of Labor's Institute of Labor and Occupational Safety and Health (referred to as ''ILOSH'') integrates image processing recognition technology to establish a hazard identification computation module for image recognition. This includes the integration of automatic mask recognition for customers entering, detection of abnormal postures of counter staff, intrusion detection for controlled areas, and the development of a recognition feature for convenience store staff uniforms. Through smart image recognition technology, automatic alerts are triggered in the event of abnormal incidents, such as unusual staff postures or unauthorized entry into controlled areas, and a reporting mechanism is established.
As image recognition and smart technology applications continue to mature, the retail industry can utilize this technology as one of the tools for monitoring workplace hazards. In the future, businesses can adjust and expand recognition capabilities based on specific needs to achieve the goal of preventing workplace hazards. This study exemplifies the real-time detection of abnormal behaviors within the operational premises of the retail industry using smart technology. Additionally, it aims to establish an automatic reporting mechanism. It strengthens the immediate emergency notification mode between businesses and law enforcement, enabling the police to respond more promptly to incidents on-site. This approach aims to reduce the time required for reporting, thereby enhancing the efficiency of incident reporting.