A Research Of Photometric Stereo 3D Reconstruction Method Based On Intelligent Terminal

Photometric Stereo is an important branch of computer vision,which restores its surface shape according to a set of images under different light.The method requires that the input image points are relatively fixed with objects,and the direction of illumination is different.According to the gray information of the image,the normal vector field and geometry of the object surface are restored.Photometric Stereo system consists of a single camera,a plurality of light sources,the classical photometric stereo technique first on a plurality of light source calibration,the calibration method is usually used as the reflection characteristics of the sphere are calculated light incident direction of three-dimensional objects known to all,a plurality of light source calibration.The shape of the surface is solved by a suitable model from the light source information and the reflection characteristic of the object.The demarcation of the light source is a tedious process,and people want to omit this step.Actually,intelligent terminals,such as smart phones and tablet computers,usually have only single light sources(flash lamps),which cannot meet the conditions of classical photometric stereoscopic technology.To solve this problem,this paper proposes an uncalibrated photometric stereo method based on intelligent terminal,this method only needs a camera and a light source shooting N amplitude images of objects from different perspectives as input data,because the image perspective is not fixed,the first from the N images in selecting a picture as positive perspective,residual N-1 image to view homography matrix,and using the homography matrix transform for image registration of N-1 images,N images are at the same angle.After obtaining the same view and different illumination N images,combined with the uncalibrated photometric stereo vision technology,we can get the high quality reconstruction results on the surface of the object without the need of calibrating the light source.It is called Uncalibrated Photometric Stereo(UPS)to recover the normal and surface photometric stereo vision from only a set of objects with unknown light and shape.In the lamb's reflection model,image gray information,normal vector,reflectivity and light source vector can form a constraint equation.The solution of the equation is an uncertain problem.Some assumptions need to be introduced to solve the model.By using the Singular Value Decomposition(SVD)and the surface integrability constraint,we can reduce the model solution to the estimation of the Generalized bas-relif(GBR)transformation matrix.There are only 3 unknown variables in the matrix.The solution of GBR parameters is the general direction of the uncalibrated stereoscopic vision problem.In this paper,the total variation regularization function and the entropy minimization function are used to jointly estimate the GBR parameters.Experiments show that the proposed method can accurately estimate the GBR model parameters.