Optical-Model-Based Analysis of Consecutive Images

摘要

Most existing algorithms for analyzing a moving scene use a 3-dimensionalgeometricmodel consisting of an object and a viewer and disregard optical processes. So, they cannot be applied to realistic scenes, such as those with light sources. This paper therefore presents a comprehensive image analysis based on a 3-dimensionalopticalmodel that includes light sources as well as the object and the viewer. The optical model provides a photometric constraint describing the geometric relationship between light sources and surface shape, and the constraint serves as the basis for a new algorithm that estimates optical flow, object shape, surface reflection coefficients, and illuminant direction from consecutive images. It is shown that the parameter is not just a correction term in estimating optical flow but also a photometric term having physical meaning. This is proved by showing that incident light source and surface reflection coefficients can be estimated using the parameter. The analysis can provide enough parameters for regeneration of a scene using computer-graphics techniques.