A high-accuracy single-frame 3D reconstruction method with color speckle projection for pantograph sliders

Abstract

The carbon slider of the pantograph is a crucial component in the high-speed rail power system, and regular maintenance is of great significance. 2D-based approach is susceptible to inaccuracies caused by defects, rainwater, stains, and other factors. Due to the weak texture and uneven reflectance of the slider, it is difficult to obtain accurate 3D information through a single-frame approach. To address this issue, this paper proposes a high-accuracy 3D imaging method for sliders, successfully resolving the long-standing problem of low reconstruction accuracy and limited generalization in pantograph reconstruction. Firstly, high-precision disparity maps of industrial component are generated using the binocular Gray Code phase-shifting method as the Ground Truth, and the color speckle pattern is subsequently utilized as the input for the model. Secondly, we propose a split-cost volume approach that achieves cost aggregation using only 2D decoding-encoding operations. Furthermore, disparity regression is utilized to provide the initial disparity for the GRU(Gated Recurrent Unit), thereby accelerating the convergence speed of the model. Quantitative experiments of optical standard sphere validated that the proposed method could achieve 3D reconstruction with a radius error of no more than 0.07 mm. The pantograph reconstruction results showed that the average absolute error of the carbon slider did not exceed 0.21 mm under reasonable exposure time and the working distance at which focus has been achieved, which meets the requirements of industrial inspection.