Three-dimensional surface reconstruction is an unsolved and central task in computer vision. Three-dimensional shape recovery based on the photometric stereo approach is an important task in optics. This thesis focuses on algorithms and system design of the three-dimensional scene recovery, and the main contributions are as follows: 1. We present the algorithm how to estimate the surface normal and reflectance of the scene from two-dimensional color images by a photometric stereo approach. The surface reflectance represents the color texture information without the effect of lighting. The algorithm overcomes the limit of constant albedo of shape from shading; and it need not find the corresponding point among the multiple images. 2. We present two novel approaches that integrate the normal vectors to the height map, a convolution method and an integration method combining a Gauss kernel function, which improve the iterative reconstruction method and the direct line integration approach, respectively. By the synthetic and real data, we analyze the four methods in detail, and conclude that the surface reconstructed by our approaches is more precise and the computation cost is lower. Our approaches improve the visual effect of the recovery scene greatly. 3. An automatic system that can capture the two dimensional color images under different illumination is designed, and it is easy to operate. The important parameters of the system are discussed, and we give the reference values. The surface recovery system from the color images based on a photometric stereo technique is implemented. The system consists of four steps. Firstly, the color images at the fixed viewpoint under different illumination directions are captured and preprocessed. Secondly, surface normal and reflectance are estimated by a photometric stereo approach. Thirdly, the height map is integrated from the gradient fields. Finally, three-dimensional model of the realistic object are obtained by texture mapping. The results on the real data are given and indicate that the proposed system can recover the shape and reflectance well.
修改评论