Automatic vison-based volume estimation of dump loading for real-time earthwork productivity assessment

Abstract

Accurate productivity assessment is crucial for earthwork projects and is primarily achieved by monitoring equipment like excavators and dump trucks. However, quantifying earthwork volume transported by dump trucks in real-time remains challenging. Traditional methods estimate volume by measuring load weight on a weighbridge, which is indirect and inaccurate. This paper proposes a real-time vision-based earthwork productivity assessment method based on a novel volume estimation algorithm. It first employs a multi-view stereo vision approach that integrates prior information on truck dimensions with deep learning-driven rigid point cloud registration to achieve high-accuracy reconstruction of 3D dump truck models. Subsequently, a convex hull slicing-based algorithm is applied to accurately calculate the load volume, while a deep learning transformer model recognizes truck license plates to determine cycle time and count. Validation in real-earthwork projects demonstrated a volume estimation error less than 4.7%, achieving an overall productivity assessment accuracy of 95.7%, outperforming the existing methods. These findings demonstrate the promising potential of automatic vision-based methods for volume estimation to improve the accuracy and efficiency of productivity assessment within earthwork operations.