Boom Mounted Camera Pose Tracking using Digital Twin Trained Model for Crane Safety Surveillance

A crawler crane is one of efficient and convenient equipment for lifting and moving heavy salvage in open field construction site. During crane operations, the transformation and tilt from the external environment, such as the lifted over weight object or strong winds, have a critical impact on the safety and operating ability of the crane. Especially when the boom radius increases, the lifting capacity can be exceeded, which may lead to a serious accident such as tipping over, overturning, and boom damage, which poses a risk of serious casualties. Accidents related to cranes at construction sites account for a higher rate compared to other construction machinery, and there is a strong need for safety surveillance technology. There is a need for intelligent safety techniques that can predict and prevent potential accidents that may occur by developing technology that accurately tracks the position of the crane boom. This study proposes an intelligent safety technique for detecting and preventing such risks by tracking the camera pose tracking and the boom position of cranes using a digital twin dataset for model training. In the experimental results, the camera position x, y, and z in a simulated environment are compared with the Ground Truth data. The accuracy of the proposed model is evaluated by analyzing the Mean Squared Error (MSE), R-squared, and L2 distance between the position x, y, and z coordinates derived from a 7-DOF vector estimated by the trained model and the ground truth data. The proposed method shows high performance and is expected to make a significant contribution to minimizing risks and enhancing safety for crane operations.