Photometric stereo shape correction using laser triangulation and tensor decomposition

To address the inaccuracy of three-dimensional shape recovery in photometric stereo, this paper proposes a fusion method that integrates laser triangulation with photometric stereo reconstruction. The method obtains three-dimensional coordinate points from the object surface using laser triangulation, converts pixel-coordinate reconstruction results to the camera coordinate system through linear equation solving with least squares optimization, and then applies tensor decomposition to correct the integrated three-dimensional shape from normal vectors. Unlike existing fusion approaches that rely on predefined error models, our tensor decomposition-based shape correction method adaptively extracts error patterns directly from observational data, enabling more flexible and accurate correction across objects with diverse geometric characteristics. Experimental results on both standard datasets and real objects demonstrate that our method significantly reduces reconstruction errors, with low-frequency error decreasing by 53.8% and high-frequency shape correlation improving by 25.6% compared to reference methods. The proposed approach shows superior stability and robustness, maintaining consistent performance across various object geometries where predefined error models often fail.