Estimation of environmental lighting from human face for illumination of augmented reality scenes

Official URL: http://192.168.1.20/record=b1378212 (Table of Contents)

In this thesis, we propose a method to solve a common problem in augmented reality domain; estimating light sources in an outdoor scene and lighting virtual objects accordingly. As a basis of our method we developed a framework based on estimation of environmental lighting from well defined objects, specifically human faces. The method is tuned for outdoor use, and the algorithm is further enhanced to illuminate virtual objects exposed to direct sunlight. In the first part of this thesis, we propose a novel lighting estimation technique where we assume a user is looking straight to mobile devices camera. This technique extracts information from input images to calculate possible light sources to pass to the rendering stage. In the second part of this thesis, we propose a lighting model which uses the output from our lighting estimation in order to make objects appear as they are lit correctly by the sun light. This model uses a mathematical technique called Spherical Harmonics Lighting for real-time realistic rendering.