Automated Detection of Track Gauge Deviations Using Video and Depth Cameras with Machine Learning
Connor Stonestreet, Hwapyeong Song, Husnu S. Narman, Pingping Zhu, Ammar Alzarrad · 2024
Ensuring the safety and reliability of railway infrastructure is crucial for transportation systems worldwide. This paper introduces a novel approach to detecting horizontal and vertical track height deviations in railways using video and depth cameras combined with machine learning. Track gauge deviation refers to the change in track gauge values from the expected to the current value. The primary objective is to reduce the time, human labor, and costs associated with inspecting large sections of railway for track gauge deviation by automating the process with machine learning. A dataset of relevant track images is selected and augmented using techniques such as grey scaling, blurring, brightness changes, and the addition of noise. This dataset is used to train several machine learning models. Various detection strategies were developed and considered, and a combination of converting pixels to real-world measurements and utilizing depth camera data was chosen. Preliminary results from our depth camera demonstrate promising levels of accuracy for estimating track gauge deviation. This machine learning approach offers a cost-effective and efficient solution for detecting track gauge deviation, thereby maintaining the safety of our railroad infrastructure.