Three-dimensional reconstruction and segmentation of an aggregate stockpile for size and shape analyses

A team of researchers from the University of Illinois Urbana-Champaign has published a paper detailing an innovative AI-powered system for analyzing construction aggregates. The method, presented in a paper on arXiv, addresses a critical gap in construction material science. Current systems require analyzing individual, manually separated particles, which is impractical for large stockpiles at job sites. This new approach allows engineers to simply capture video or images of a pile of gravel or crushed rock using a standard smartphone camera.

The core technology leverages computer vision techniques, specifically Structure-from-Motion (SfM), to reconstruct the surface of the stockpile into a detailed 3D point cloud—a massive dataset of spatial coordinates. A custom 3D segmentation algorithm then processes this cloud to digitally separate and extract thousands of individual aggregate particles from the tangled mass. The system can subsequently calculate key morphological properties like size distribution and shape (angularity, flatness), which directly influence the stiffness, strength, and durability of road bases and other geotechnical layers.

This research, presented for the 20th International Conference on Soil Mechanics and Geotechnical Engineering, demonstrates a shift from lab-bound, sample-based testing to potential real-time, field-based evaluation. The preliminary results show the future potential for a convenient and affordable QA/QC tool. By providing immediate, objective data on material properties, it could lead to more consistent construction quality, better-informed material sourcing, and optimized pavement design, all from a device already in an engineer's pocket.