Three-dimensional Image Depth Estimation Based On Light Field Photography

People record a variety of information through the cameras, while traditional digital cameras can only record two-dimension information, they are lack of perception of the depth information. How to perceive three-dimensional world’s depth info better becomes the focused research in recent years. Light field cameras’ appearance solves this problem, they can achieve the goal of refocusing after take photograph and get any depth of field image. The micro-camera array structure’s advantage is easy to carry, close to the actual application and has good prospects. This structure contains pairs of binocular cameras, binocular vision is a classical method of calculating depth information, has the advantages of high precision and low system cost. Therefore, the content of this paper is not only the key steps of three-dimensional reconstruction, but also the key technology of accomplishing digital refocusing in light field imaging. In this paper, we do the researches and explorations through a binocular vision model in two main key steps:camera calibration and stereo matching. Specific research contents are as follows:1. Camera Calibration:Introduced the common methods of camera calibration and compared various types of the method, selected the calibration method based on planar template and expounded its relevant theory in details. The specific steps of calibration is: Used the left and right cameras to take pictures with board calibration template ten times at the same time, got the internal parameters of left and right camera after corner detection and single camera calibration in MATLAB. Then carried the binocular stereo calibration work, got the external parameters of the camera such as rotation matrix and translation vector, and the internal parameters after corrected.2. Stereo Matching:Introduced stereo matching algorithm and constraint conditions used in matching, using the theory of epipolar geometry and obtained images to finish the binocular stereo correction job. Combined with the calibration results in MATLAB, achieved the epipolar correction and stereo matching steps in OPENCV. Used the SGBM stereo matching algorithm to generate the disparity map, and estimated the depth information according to the triangulation principle. Experiments set up the mixed programming platform which based on MATLAB and OPENCV, experiments were compiled with the software environment Visual Studio2008and Opencv2.4.3.