Research On Photometric Stereo Algorithm Based On Near-field Lighting

Photometric stereo (PS) is an important method of three-dimensional scanning technology, and also an important research field in computer vision. The resconstruction system based on PS algorithm has simple and compact structure as well as low price and high performance of reconstruction results. Therefore, PS algorithm attracts more and more attention from academia and industry and has been applied widely in the industrial, medical and cultural fields and so on. The lighting direction and image brightness of the surface are two important factors, which affect the accuracy of surface normals directly. For near-field lighting, the relationship between light rays and a surface becomes complex, which makes light source calibration become more complicated and difficult. Besides, the images for solving the surface normal usually need to obey certain kind of intensity distribution, which are different from real images. To achieve realistic surface reconstruction results under near-field lighting conditions, this dissertation took deep and careful research into light source calibration and proposed a series of innovative and improved algorithms.At the beginning of the dissertation, photometric stereo algorithm was reviewed in details, including the content, the trends and the achievement.Then, the implementation process of photometric stereo algorithms was discussed, and the system and its calibration procedure were described in detail for algorithm verification.Reflectance model is not only used in the process of normal vector estimation of the surface, but also used to obtain lighting information. Using an accurate reflectance model, the direction of light rays can be calibrated precisely. A novel reflectance model, i.e. a dual-Lambert reflectance model was proposed, where the dual-Lambert refered that the distribution of illuminating radiance and the reflective characteristics of an object surface both were Lambertian characteristics. The experimental results showed that the proposed dual-Lambert model can more accurately characterize the reflectance characteristics of an object surface compared with traditional Lambertian model, and give a better calibration results. A light source that is only considered as specific shaped geometry regardless of the characteristics of the light itself is called an ideal light source model. According to the size and the shape of an light, the model is divided into three types, i.e. infinity point light source, near point light source and area light source. For the construction system under the condition of near-field lighting, we compared the reconstructing results using different lighting models and provided a theoretical and experimental basis for the selection of reasonable model. The distribution pattern of an area light source was extended so that any pattern of an area light source can be used in PS.Existing near ideal point light source calibration methods relys solely on the spatial geometry or the established light illumination model, ignoring the effect of the specific characteristics of the light source. As a result, for the same position of two different types of light source, the calibration results are different. Based on the proposed double-Lambert illumination model, three calibration algorithms incorporating the characteristics of a light source were presented,. Experimental results showed that the proposed algorithms compared with existing algorithms, obtained more accurate sources of information, and can effectively improve the reconstruction results.Heavy computing is needed in the process of calibration, which makes the entire calibration process become very time-consuming. Especially, for the model fitting algorithm using image information, the calibration process often requires repeated iterations, which makes the entire calibration process become very time-consuming. In addition, due to the interference of various factors, the calibration results is often uncertain. For the above, a fast calibration algorithm for near point sources was presented, in which the position of any light source can be estimated fast and accurately using only a single image. Experimental results showed that the algorithm can achieve real-time source calibration, along with a high calibration accuracy.Errors in the manufacturing process of illuminators, installation process, and changes in the spectral response of optical materials as well as random choice can result in significant difference in radiance, which make image brightness distribution inconsistent and affect the results of the reconstruction. Most existing light radiance correction algorithms focused on the parallel ray’s model and overlooked the case of near point source model. A correction algorithm based on the near point sources to compensate the radiance was proposed. Experimental results showed that the algorithm can effectively compensate for light radiance and improved reconstruction results.