Get started
Work

A Brand New Single-shot Structured-light Sensor for Railway Infrastructure 2D&3D Visual Inspection

Year: 2025
Type: conference-paper
Abstract: Machine vision detection technology serves as a crucial technical means for identifying structural damage in railway infrastructure. Currently, the railway infrastructure machine vision inspection system primarily comprises two components: two-dimensional (2D) visible light imaging and three-dimensional (3D) shape measurement. While 2D visible light imaging effectively captures color and texture features, it lacks depth information. Conversely, 3D visual data characterizes shape features but lacks texture details. The combined detection method using 2D cameras and 3D sensors faces challenges due to parallax between the two modal sensor data, making accurate registration difficult. To address these limitations, this paper presents the development and design of a novel single-eye optical sensor based on line structured light imaging. This sensor employs a high-speed CMOS chip to capture multi-frame laser stripe images within a single exposure period. Utilizing Gaussian dynamic frequency domain filtering, noise in the multi-frame images is suppressed, achieving a balance between overexposure and underexposure within the same imaging area. By fusing the results from multiple image light stripe regions, the sensor synchronously outputs both 2D grayscale images and 3D shape data of the measured target. A prototype of this sensor was developed and successfully applied to detect defects in rail surfaces, fastener structures, ballastless track slabs, ballasted track ballast, and other railway infrastructure components. Experimental results demonstrate that the 2D and 3D images output by the sensor are accurately registered with zero error. The 2D images exhibit comparable texture detail representation capabilities to 2D visible light CCD imaging, while the 3D measurement accuracy in the Z-axis direction achieves ±0.1 mm.
DOI: 10.58286/31065
Authors: Haoran Wang, Haoran Song, Le Wang, Qiang Han, Jun Li, Peng Dai
Language: English
Open Access status: gold
OpenAlex ID: W4409730575
FWCI:
Cites:
Cited by:
Related to: 10