Calibration And Depth Estimation For Light Field Cameras

Light field camera which as a new kind of multiview camera can record the intensity and direction information in space simultaneously.We can use these information to complete depth estimation,which is of great importance for visual tasks such as 3D reconstruction.However,the position of micro-lens array center changes and the pixel of captured image shifts because of fabrication process and manufacturing technique.At same time,the existing depth estimation research are scattered and not systematic.Also,the effect of illumination variation and different complexity scenes on the performance of depth estimation has not been carried out.Therefore,we focus on calibration and depth estimation of light field.The main work of this thesis is listed as follows:(1)We summarize the background and research status of light field calibration and depth estimation method,then introduce the basic theory of acquisition,parameter representation,Epipolar Plane Image,multiview image and refocusing of light field.(2)We design a new local search micro-lens array center calibration method to reduce the influence of main lens installation error and diffuse reflection light uniformity.It can improve the calibration accuracy.Then we use color correct and refocus to verify the robustness and effectiveness of our algorithm.(3)We acquire the light field white image,raw data,focal stacks,all-focus image and multiview to construct a more abundant and varied dataset with Lytro,which provides data support for light field depth estimation.(4)We review the existing depth estimation method at home and abroad,then discuss the characteristics of typical algorithms.Further,we classify them into depth based on Epipolar Plane Image,multiview and refocusing three methods.Also,we analyze the effects of illumination changes and scene types on the performance of depth estimation,then point out the future direction of light field research.