Automatic Safety Monitoring of Construction Hazard Working Zone: A Semantic Segmentation based Deep Learning Approach

Automatic Safety Monitoring of Construction Hazard Working Zone: A Semantic Segmentation based Deep Learning Approach

연구 분야: Verification

학회: ICAL 2020: Proceedings of the 2020 the 7th International Conference on Automation and Logistics (ICAL)

This paper presents an application of Semantic Segmentation-based Deep Learning (DL) technique to achieve real-time safety monitoring of construction hazard working zone, so that the unsafe situation can identified timely to reduce safety risks. Two different Convolutional Neural Network (CNN) based Deep Learning (DL) techniques were adopted for worker identification, and target working zoning, including Faster R-CNN, DeepLab v3+. A sample hazard working zone near building elevator shaft is adopted for case study. The opening of safety fence as well as the working man nearby is identified as a target hazard scenario to be detected. From both of the results of lab and in-situ testing, it is found that all performance indexes including the Recall and Precision during the training process in lab and the Cleanness and Correctness obtained on site surpassed the 95% high criterion values. It is therefore concluded that the proposed method provides the construction safety personnel an effective tool to monitor the risk and prevent the accident for the construction workers in hazard working zones.