Photorealistic 3D Model Reconstruction Based on Surface Reflectance and Illuminant Direction

Description

There is increasing demand for high fidelity 3D models of real objects in various areas. Image-based model reconstruction makes use of the image sequence of the target object acquired by an off-the-shelf camera. This approach has the advantages of low equipment cost, flexible operation, and the quality of the reconstructed model can be improved in line with advances in camera and computing technologies. But generating a 3D model with true realism is a challenging problem. Existing methods generally impose a constraint on the type of object surface that can be reliably modelled. They also make assumptions on the illumination that are not realistic. The reconstructed models therefore demand a lot of post-processing effort before they can be used in practical applications. This project proposes a novel volumetric model reconstruction algorithm that can tackle objects exhibiting both diffuse and specular reflections. The identified diffuse surfaces are used to determine the illuminant orientation. Both volumetric model reconstruction and illuminant estimation are refined in a Gauss-Seidel approach. The model reconstruction algorithm can generate models for a wide range of objects in high realism (in terms of geometric accuracy and visual appearance) under a less constrained environment, and it is still computationally manageable.