3D reconstruction of specular surface by combined binocular vision and zonal wavefront reconstruction

This paper proposes a framework for the three-dimensional (3D) reconstruction of specular surfaces. The framework begins with a modified reconstruction method that uses gradient descent to accurately search for the correspondence matches in the binocular vision system. An analysis of the binocular vision system shows that the reconstruction result is not necessarily unique and the reconstruction accuracy is significantly compromised by a bilinear interpolation error. An error-based selection algorithm is proposed to select the correct result. To improve the reconstruction accuracy, binocular vision and zonal wavefront reconstruction are incorporated. The reconstruction result from the binocular vision system is taken as a reference surface for slope estimation. The measured slope is then integrated into a height map by the zonal wavefront reconstruction. The overall reconstruction process provides an accurate reconstruction result without the need for prior surface information. Simulations and experiments were conducted to verify the feasibility of the proposed algorithms. A relative error of 0.59% or less is experimentally recorded in the step-size measurement of a stair-like object.