Worker Action Recognition in High Density Conditions based on Edge Device – Cloud Infrastructure

Construction managers face increasing workloads and time-consuming activities. However, there has been limited application of modern technology like deep learning to improve efficiency on construction sites. This paper focuses on the capabilities of plain vision using YOLO, StrongSORT, 3D CNN, pose estimation and machine learning models for detecting, tracking, and recognizing worker actions. Emphasis is placed on model accuracy for dense and diverse objects per frame, along with proposed solutions for improved YOLO results. Additionally, an adaptable application is developed using a cloud-edge device model, comprising AI and user management parts, to handle high AI workloads in rapidly changing construction environments.