Automatic 3D railroad alignment detection using modified Hough transform

In railroad infrastructure, timely and accurate maintenance is essential to ensure both safety and operational efficiency. Rail alignment is a critical element for constructing a reliable digital model of railroads. However, conventional rail alignment detection often relies on manual processes that are subjective and prone to error. With the advent of digital modeling technologies, there is an opportunity to improve the efficiency and accuracy of railroad maintenance. In this study, we propose an automated method for detecting railroad alignment directly from 3D point cloud data using deep learning and computer vision. The method was validated on the Osong railroad test track, achieving an average RMSE of 3.57 mm in rail alignment detection. The proposed approach significantly reduces the time required to construct digital railroad models and enhances the efficiency of railroad maintenance.