Target Depth Estimation Based On Light Field Information

As a revolutionary optical acquisition system,light field camera has many advantages over other methods in depth estimation.Based on the actual effect of depth estimation,this paper studies the method of scene target depth estimation based on light field information,and uses super-resolution reconstruction algorithm to improve the effect of depth estimation.First of all,we summarize the light field theory.We expounds several method to get light field data,studies the decoding process of existing light field cameras and preprocessing methods,and processes the field collected light field data according to the research.Then,this paper studies the algorithm to estimate depth.We design a network structure to learn the importance of different views of the input light field image through the view selection module,and combines with the new operator evolution to better estimate the scene depth.Finally,in order to solve the problem of insufficient resolution of depth estimation results of light field image,this paper studies the existing image superresolution algorithm,and combined with the relevant knowledge of transfer learning,optimizes the super-resolution reconstruction method based on feedback mechanism,completes the task migration from color image to depth image,and improves the spatial resolution of depth image.The experimental results show that,in the depth estimation part,for the estimation of public test data sets,the method used in this paper estimates the scene target depth better.Compared with other methods,the depth estimation result of this method is more accurate,and the performance in mean square error and bad pixel index is better.In the estimation of field data,this method performs better for the near scene target,but worse for the far scene target.In the part of super-resolution reconstruction of depth map,the method used in this paper can effectively improve the resolution of depth results.Compared with other methods,the performance of this method is better,the reconstruction results are closer to the original image,and the subjective evaluation and peak signal-to-noise ratio are better than other methods.