Joint optimisation model for light source position based on a combined target

Abstract

Photometric stereo vision, as a non-contact three-dimensional imaging technology, can achieve fast and high-precision reconstruction of normal vectors and three-dimensional morphology of object surfaces by resolving the intensity changes under different angles of illumination. However, the existing near-field photometric stereo vision system is limited by the complexity of the light source calibration process and the error accumulation, which seriously restricts the expansion of the application of this technology. This paper addresses the problem of accurately determining the centre of a spherical mirror target combined with a planar calibration board. We proposed a joint optimization model that simultaneously estimates both the light source positions and the sphere centre coordinates. The model defines a unified objective function to guide the optimization. The calibration is implemented using the Ceres Solver, based on near-field photometric stereo principles. Compared with the pre-optimization results, the error in the estimated sphere centre is reduced from 0.541 mm to 0.228 mm.