Earthmoving is one of the key activities in most heavy civil construction projects. The dump truck is one primary construction vehicle for earthmoving. Two popular approaches are currently used to estimate earthmoving volume by trucks, i.e., manually counting the number of loaded trucks and weighing loaded trucks on a scale station. Considering both methods are either error-prone, time-consuming, or costly, this study aims to estimate different earth volumes in dump trucks from a single image using the machine learning approach. By establishing a pre-trained deep learning neural network from 3663 images with sixteen different volumes of the earth using a scaled dump truck model, the proposed approach is tested to estimate the truckload in a quantitative manner in real-time. Another 1221 images are used for verification in six case combinations out of the sixteen different volumes. The preliminary results show that the classification accuracy by using the pre- trained network is 100% if the volume gap between adjacent classes is more than 5%, while 76.67% if the volume gap is 1%. The preliminary test results show a great potential that the proposed methods could be applied to the field and provide a fast and accurate estimate of truckload with minimal cost.

No abstract