Robust contact-point detection from pantograph-catenary infrared images by employing horizontal-vertical enhancement operator

Abstract

The pantograph-catenary geometric parameters are important indicators for real-time monitoring the normal operation of electric multiple unit (EMU) trains, the stable and accurate contact-point between pantograph and catenary is a key factor to measure geometric parameters. However, due to the interference of complex backgrounds (e.g, cloud, cross-bridge, curved contact wire, and etc.), it challenges for image information processing technology to accurately and robustly detect contact-point from infrared images. In this paper, an efficient contact-point detection (CPD) scheme to detect contact-points from these complex infrared images, including the following three key components: Firstly, the proposed horizontal-vertical enhancement (HVE) approach separates a single input infrared image into a horizontal image layer (HIL) and a vertical image layer (VIL). Secondly, the contact-wire model is used for the point detection of the contact wire from HIL while the pantograph model is used for the point detection of the pantograph from VIL. Finally, the contact wire and the pantograph are detected by an improved RANSAC strategy and the contact point is positioned by using these two components. The quantitative and qualitative results validate the effectiveness of the proposed scheme. Moreover, the proposed method presents its robustness and efficiency for contact-point detection at a over speed of 164 fps, 0.48 average pixel error, and 99.6% average accuracy in two datasets (12,000 frames), which is very beneficial for its extensive application.