Railway Engineering

Abstract

This study introduces an obstacle and curve detection system based on the Hough Transform, specifically designed for small drones that capture imagery while flying over railway tracks. Drones have previously been utilized for monitoring railway infrastructure and detecting potential faults along the track. On the other hand, artificial intelligence has been applied in railway systems for detecting both obstacles and rail defects. However, these technologies are still open to development. Therefore, to enhance reliability and provide a supportive or redundant mechanism, it remains important to develop obstacle detection algorithms based on image processing, independently of AI-based systems. The proposed approach suggests monitoring the section of the Hough transform along the sinusoid of the track’s vanishing point to detect of obstacles located on the rails. By this way, the two-dimensional Hough transform information is reduced to a one-dimensional representation, in which all Hough peaks can be observed simultaneously. This reduction leads to a less complex algorithm suitable for implementation on small drones. For curve detection, the image is divided into smaller horizontal segments, and the movement of Hough peaks is analyzed accordingly. To evaluate the performance of the proposed solution, railway animation videos simulating real scenarios where drones record video while flying over tracks have been generated. Test results demonstrate that the proposed method effectively detects obstacles on the railway tracks.