Research And Implementation Of3D Reconstruction Based On Photometric Stereo

Three-dimensional modeling of objects based on computer vision belongs to non-contact three-dimensional measurement. Because its operation is very simple and it has the advantage that it doesn’t need to contact with objects which the other three-dimensional reconstruction never have, it is widely used in the Internet applications, reverse engineering, industrial inspection fields and so on. In addition, the development of film special effects and face recognition depends on the research and application of three-dimensional reconstruction techniques in the computer vision increasingly. Therefore, from the aspects of visual physiology and engineering application, the researches of three-dimensional reconstruction based on computer vision are very significant.Based on the analysis of traditional methods, a relatively new method of three-dimensional reconstruction photometric stereo method is applied to reconstruct three-dimension of objects in this thesis. Based on the relevant theory of physical optics, the relationship between the surface towards and the surface brightness is gained by analyzing the reflection model of object surface, which is the basic principle of the photometric stereo. The main steps of this method are that the normal vector distribution of the features surface is calculated through multiple images firstly. And then, the three-dimensional information is restored by the obtained normal vector. The pretreatment of images is done by ways of the camera response curve and median filtering to eliminate the brightness bias and noise impact on the subsequent treatment. Afterwards, the specific algorithm process of calculating the surface normal vector with multi-light photometric stereo method is given. Besides, the interference is eliminated in the presence of mirror reflected light by using multi-light photometric stereo method. Moreover, after the normal vector field is gotten, the triangular mesh method based on discrete model is applied to reconstruct the surface and the situation of the algorithm instability solution is analyzed. At last, the algorithm is improved by adding pre-defined conditions.Firstly, the current development status of computer vision is presented and several common three-dimensional reconstruction methods based on images are discussed. Secondly, every aspects of photometric stereo is analyzed and studied systematically. And the algorithm of the surface reconstruction phase is improved. Finally, the performance of every aspects of the algorithm is verified through a large number of experiments. Experiments results show that the algorithm has good performance on the aspects of accuracy and reconstruction effects.