Research And Application Of Depth Information Estimation Method Based On Light Field Imaging

Compared with the traditional imaging method,the imaging method based on the light field not only records the position information of the light radiation in the scene,but also records the direction information of the light radiation.The light field image"single exposure and multiple views,first shooting and then focusing"feature opens up new ideas for passive depth estimation methods,and has important research significance.However,how to make full use of the four-dimensional light field information while solving the problems of narrow light field image baseline and large occlusion area depth estimation error is an important topic of this research.The main work of this thesis includes:1)Propose a matching cost calculation method based on sparse AD-Census and cross multi-viewpoints.In order to obtain a matching cost with higher accuracy,this thesis introduces AD-Census,which has the advantages of higher robustness to noise and light intensity,into the stereo matching method of the light field.At the same time,in order to ensure the running time of the algorithm,this thesis proposes to optimize the calculation of AD-Census through sparse matrix,and reduce the amount of calculation of matching cost by selecting the views based on cross multi-viewpoints.Experiments show that this method has a smaller estimation error than other light field cost calculation methods.2)Propose a global depth optimization method based on the light field occlusion model.In order to reduce the depth estimation error caused by the occlusion problem to the light field depth estimation method,this thesis improves the energy function used to obtain the disparity map.Specifically,on the one hand,the matching cost based on AD-Census and the depth cues based on the light field occlusion model are confidently fused to optimize the E_data(f)term of the energy function;on the other hand,the E_smooth(f)term of the energy function is further improved by occlusion detection.Experiments show that this method can effectively improve the accuracy of the estimated depth of the edge area of the object.3)Build a three-dimensional reconstruction system for the light field.According to the basic principle of light field imaging and the depth estimation method based on AD-Census and light field occlusion model,this thesis realizes a light field 3D reconstruction system.The system can provide functions such as light field image processing,3D model reconstruction and 3D model management,so that users can easily and quickly reconstruct 3D models based on light field images.