A 3-D Feature-Enhanced Image-Based Deformation Detection Framework for Railway Pantograph Horns

Deformation detection is a critical element of industrial defect identification, exerting significant influence across various facets of industrial production, operation, and management. This paper defines a novel structure termed as the MSCCS (Multiple Side-by-side Coplanar Curve Structure) and introduces a deformation detection methodology predicated on two-dimensional imaging. This methodology is further applied to railway pantograph horns. Initially, a framework is delineated for detecting MSCCS deformations, which encompasses target edge extraction, relative 3D position normalization transformation (R3DPNT), curve registration, and deformation evaluation. Specifically, a two-dimensional image-based R3DPNT approach and a proficient curve registration process are elucidated. Furthermore, an edge extraction network tailored for pantograph horns along with deformation evaluation metrics are proposed. Experimental results indicate that R3DPNT demonstrates efficacy, and the proposed registration technique surpasses the Iterative Closest Point (ICP) method relative to the Hausdorff distance and the square of Euclidean distance. The proposed edge extraction network attains 0.8969 IMP, 0.8265 ODS, 0.8298 OIS, and 0.8841 AP with a parameter size of 3.425 M and a frame per second (FPS) of 41.4. The advanced deformation detection methodology can be efficiently deployed for online detection of pantograph horn deformations on active railway lines.