Stereo Vision–Based Icicle Detection for Tunnel Deicing

Abstract

Icicle damage is common in tunnels in cold regions, posing a major threat to the operational safety of railway trains. To address the many drawbaks in existing deicing methods, this paper proposes the design of a tunnel deicing vehicle with an icicle detection and positioning system. Firstly, a tunnel deicing vehicle structure is designed for the tunnel scenario in which the deicing vehicle operates. Moreover, icicle detection experiments are conducted based on the You Only Look Once (YOLO) deep-learning method. In addition, ice positioning experiments are carried out based on the stereo vision, with the relative error within 2.2%, proving the feasibility of the detection and the positioning system. The combination of the deicing vehicle and the binocular vision system enables the detection, localization, and cleaning of the icicles. This deicing process reduces manual intervention, improves the efficiency and quality of deicing operations, and ensures the safety of railway operations.