Ballast plays an important role in supporting the railroad track structure against repeated loading. Ballast degradation causes poor drainage, lateral instability, and excessive settlement, and may lead to service interruptions and safety concerns. Therefore, utilizing efficient methods for evaluating ballast degradation ensures safe railroad operations. Traditional visual inspection is subjective and field sampling followed by laboratory sieve analysis is labor intensive. 2D image-based computer vision approaches have been effectively incorporated into ballast size inspection to enhance accuracy and robustness. Ballast shape properties, including Angularity Index (AI) and Flat and Elongated Ratio (FER), also have a significant impact on the performance of the ballast layer. This project will integrate a novel 2D image-based deep learning algorithm on mobile devices to analyze ballast size and shape properties in the field in real-time.
Considering the limitations of lacking spatial information for 2D imaging approaches, there is also a demand to expand the inspection process to the 3D domain. Building a detailed spatial profile of the testing spot benefits the evaluation of 3D ballast size and shape properties, which are more valuable in practical use. This project will establish and test a 3D field ballast data acquisition system, develop both 3D ballast particle segmentation and completion algorithms, design and implement the 3D size and shape analysis workflow, and eventually encapsulate these in user-friendly software.