Zero-shot High-risk Situation Detection Based on Semantic Segmentation and Pose Estimation Using Fixed-point Cameras at Construction Sites

In this study, we propose a zero-shot high-risk situation detection method based on semantic segmentation and pose estimation using fixed-point cameras at construction sites. Our goal is to build a model that can help to prevent construction site accidents. The proposed method combines a pre-trained object detection model with a pose estimation model and uses semantic segmentation to perform highly accurate vehicle detection. This allows the detection of two high-risk situations, where the person is on the roadway and the person does not recognize the vehicle. Through the reporting of high-risk situations detected by the proposed method, it is expected to promote safety awareness among workers and managers. Finally, the effectiveness of the proposed zero-shot high-risk situation detection method is verified through experiments using videos captured at the entrances of construction sites.