Explicit and Implicit Fusion Model-Based Digital-Twin Virtual Scene Construction Under Dynamic and Irregular Engineering Scene

In recent years, digital-twin technology has gained significant attention in the field of engineering construction due to its potential in project planning, execution, and management. However, traditional digital-twin virtual scene construction methods struggle to handle the highly dynamic and irregular nature of engineering construction scenes. To address this issue, this article proposes a digital-twin solution for engineering construction scenes by integrating explicit and implicit models. This solution uses visual sensor information from the construction site to divide the scene into highly dynamic regular objects and low dynamic complex backgrounds, with the former being represented by explicit models and the latter by introducing modified neural radiance field (NeRF) implicit models. Additionally, a fusion rendering method is designed to enable the integration of explicit and implicit models. To achieve real-time mapping of target positions in the physical scene, this article proposes a visual target localization algorithm based on feature maps and AprilTag, allowing real-time mapping of targets from the physical scene to the digital-twin scene. The experimental results show that the digital-twin model constructed for the engineering scene in this article demonstrates impressive realism, with peak signal-to-noise ratio (PSNR) values of 24.25 and 27.71 for indoor and outdoor building scenes, respectively. Additionally, the absolute position error (APE) root-mean-square error (RMSE) of our tracking trajectory error is around 0.032, and we ensure the efficiency of mapping the target pose between the physical scene and the digital-twin virtual scene reaches over 30 frames/s. The experiments fully demonstrate the effectiveness of our method, emphasizing its revolutionary potential for digital-twin applications in engineering construction.