인공지능 기반 전동차 집전장치 수평 자세 추정 방법

초록

This study proposes a vision-based horizontal pose diagnosis algorithm for the automatic inspection of pantographs in urban railway vehicles. Traditional visual inspection methods are prone to inconsistency due to inspectors’ levels of expertise, and are often time-consuming. To overcome these limitations, the proposed method detects the pantograph in an image and estimates key joint coordinates to quantitatively assess its horizontal pose. Horizontal inclinations are calculated for three line segments formed by pairing six upper joints. Using reference angles and allowable deviations derived from normal-state data, the system determines abnormalities in a quantitative manner. Experimental results demonstrate that the proposed method achieves a horizontal angle estimation accuracy within ±1 degree. This approach addresses the subjectivity of manual inspections and can serve as a core technology for real-time diagnostic systems aimed at automation and fault prevention. Furthermore, it is expected to be scalable to various vehicle types and structural configurations.

키워드