Research And Implementation Of 2D-3D Based On Automatic Depth Map Generation Algorithm

As people enter the age of intelligence,depth maps are used in unmanned driving,intelligent monitoring,3D resources and other fields.Since the shooting time of 3D movies is very long and costly,the 3D video resources produced each year are still rare compared to 2D video resources.Therefore,scholars began to use depth estimation algorithms to convert 2D resources into 3D forms.This paper implements a variety of fully automated methods for extracting depth maps.The main work can be summarized as follows:(1)Optimized the scene clustering depth map generation algorithm and integrated scene clustering and linear perspective to generate depth map.The Konrad scene clustering method is compared with the depth map generation method based on scene search and linear perspective fusion.Experimental results show that the improved algorithm on the peak signal-to-noise ratio index when the target image has linear perspective depth clues,and the overall depth of the image is more obvious.In addition,two methods based on Euclidean distance and visual words are used to search the approximate scene of the target image,respectively solving the two problems that are similar to the target image or less similar to the target image in the database.(2)Proposed a depth map generation algorithm based on convolutional neural network.Aiming at the problem that the deeper the network is,the more information is lost and the generated depth map is inconsistent with the original image size,a convolutional network model based on ResNet residual unit and deconvolution operation was built.Finally,through comparative analysis of experiments,it is concluded that the depth map generation algorithm based on convolutional neural network has the most extensive application scope,the best peak signal-to-noise ratio performance,and the generated depth map has obvious details and overall changes.(3)Completed the experiment of the whole process of 2D-3D conversion.Aiming at the problem of voids in virtual viewpoint images,the operation of image dilation and image erosion was adopted to obtain ideal 3D effect images.