Field imaging framework for morphological characterization of aggregates with computer vision: Algorithms and applications

A new PhD thesis by Haohang Huang presents a comprehensive computer vision framework designed to revolutionize how the construction industry analyzes aggregates like sand, gravel, and crushed stone. The research, published on arXiv (2603.03654), tackles the limitations of current methods that rely on manual inspection or controlled lab settings. Huang's work introduces a multi-scenario solution that works directly in the field, addressing the core challenge of automating the morphological characterization of irregular materials in real-world, messy environments like stockpiles.

The technical breakthrough is the integrated 3D Reconstruction-Segmentation-Completion (RSC-3D) approach. First, a high-fidelity 3D particle library is built from reconstructed aggregate samples. From this, two synthetic datasets are generated for machine learning: one for 3D instance segmentation of stockpiles and another of partial-complete shape pairs for training a shape completion network. This allows the system to not only segment individual particles from a pile but also predict their hidden geometries. The framework was validated on real stockpiles, demonstrating accurate capture and prediction of unseen sides, moving the industry toward fully automated, data-driven material assessment.