Evaluation of Railroad Ballast Field Degradation Using an Image Analysis Approach

Abstract

Identifying the level of ballast degradation generally involves ballast sampling and mechanical sieve analyses in the laboratory, which can be time consuming, laborious and costly. As an automated alternative, image processing techniques has the potential to directly and objectively assess ballast condition and degradation levels from high resolution images of ballast layers captured in the field or reproduced in the laboratory. This paper presents the development stages and implementation of an innovative image processing method for assessing the degradation levels of ballast using ballast cross section images collected in the field and also reproduced in the lab. Advanced image enhancement methods, including gamma adjustment, histogram equalization, and bi-lateral image filtering, combined with image segmentation techniques such as watershed algorithm and image thresholding, were used to successfully extract size and shape properties of individual ballast particles as a mean to quantify the level of ballast degradation. In order to capture images of the ballast layers in the field, a detailed procedure was developed to ensure the resulting images captured would perform consistently and accurately when processed with the machine vision algorithms. Rapid imaging of a large quantity of ballast samples was needed for producing ground truth data to be used as input into the machine vision algorithms. The results of this study showed that the images captured in the field and the images captured in the lab from the corresponding collected ballast samples looked quite different. This confirmed that a robust image processing algorithm which can be linked to indices based on sieve analysis methods needs to be adjusted/trained from the images and samples collected in the field. The findings of this ballast field and lab imaging study showed promising future potential of the described image processing technique for replacing the tedious and time consuming ballast sampling and sieve analysis processes for evaluating ballast degradation.