Depth Flow Estimation Of Light Field Based On Variational Method And Occlusion Characteristic

The 4D light field data captured by light field cameras including spatial and angular information of rays can be used for scene depth estimation.Depth estimation of light field is widely used in machine vision and 3D reconstruction.Affected by the complexity of scene,such as noise,smooth texture and occlusion area,scene depth estimation usually faces the problems of low accuracy and poor robustness.In this paper,light field flow estimation algorithm based on variational method and depth estimation of light field algorithm based on occlusion detection are proposed by using 4D light field data and the corresponding relationship between depth and parallax.In this paper,the problem of depth estimation.of light field is transformed into the problem of disparity between views by using the relationship between disparity and depth,and the disparity or light field flow of sub-aperture images is calculated by using 4D light field data.Firstly,in chapter 3,a variational method is proposed to solve the light field flow based on the classical optical flow algorithm.Different from the existing depth estimation algorithms,the proposed algorithm studies the depth estimation of light field from different angles.According to the principle of digital refocusing,parallax contained in 4D light field data is directly introduced into variational model of light field flow.The gradient constant hypothesis and global smooth hypothesis are integrated on the basis of the traditional variational optical-flow framework.The robust penalty function is used to eliminate outliers of data and smooth,and the improved image pyramid based on median filtering is used for multi-resolution computation.Thus,a robust field flow estimation algorithm with better performance is obtained.Then,in Chapter 4,based on the idea of light field flow algorithm of 4D light field data,an improved light field flow variational model is constructed by using the 2D representation of sub-aperture images.By minimizing on field flow equation,the corresponding Lagrange-Euler equation is calculated by SOR iteration.Then,based on the improved variational model,sub-aperture images are used to estimate field flow from local to global.Finally,the mean value optimization and weighted median filtering are used to optimize the global field flow.The above researchs has not yet solved the occlusion problem,and occlusion detection is very important for accurate depth estimation of optical field.Therefore,in chapter 5,the paper focuses on the geometric complementary occlusion model of sub-aperture image array,and combines this occlusion model with depth estimation of optical field as the final optical field flow model.The optical field flow model proposed in this paper includes optical field flow estimation and occlusion detection.More specifically,first of all,the sub-pixel optical flow of any two sub-aperture images is estimated by using the phase-shift theory,and the cost of constructing the optical field flow is also estimated.Then,the occlusion pixels are detected by the consistency of the front and rear optical flow.Finally,a light field model based on occlusion detection is constructed and depth estimation of light field is carried out.Experimental results show that the depth estimation algorithm proposed in this paper greatly reduces the computation error of occlusion region.Compared with the existing algorithms,the algorithm proposed in this paper has better robustness and computational accuracy.