Abstract

Construction and mining require high levels of planning and productivity. Planning and management of operational targets and performance requires taking into account several variables, among which the amount of moved, loaded and unloaded material is one the main ones. This motivates the development of an approach for the estimation of the volume of material transported in field operations by loader trucks. The proposed methodology employs a three-dimensional point cloud representation of the volume and is tested using a skid-steer loader. The volume estimation process involves the segmentation of the loader's bucket from the raw point cloud using machine learning techniques. The segmented point cloud is associated with a pre-constructed reference model of the empty bucket. The volume is computed from a surface model built using the alpha shape algorithm applied to the Delaunay triangulation of the segmented point cloud. Several field experiments were carried out using a stereo camera for five different volumes of material and the bucket at different heights. A 95% accuracy was obtained on average. The encouraging results suggest that effective volume estimation for different bucket types would be possible using the proposed methodology.