Image-Based 3D Reconstructions for Stockpile Volume Measurement

This paper presents a vision system capable of providing an accurate estimate for the volume of stockpiles using 2D images. By imaging a stockpile from multiple viewpoints with a two-camera system, a 3D point cloud of known scale is generated using image-based 3D reconstructions. With a small amount of user interaction, the points that correspond to the surface of the stockpile are extracted and used to measure the volume. The hardware required for the two-camera system is light enough to be carried by a small unmanned aircraft and is a cheaper alternative to LiDAR while still being competitive in terms of accuracy. When testing the volume measurement algorithm on a stockpile of known volume, the measured output was found to have less than a 3% error from the ground truth.