Deep Convolutional Network Based 3D Target Modeling for Multi-view Video

This paper studies the 3D target modeling method under multi-view video based on deep convolutional network.Through the detailed exposition of the basic theory of 3D target modeling technology and the complete derivation of non-uniform rational B spline curve, this paper establishes technical support such as camera coordinate system for the generation of 3D target model.According to the basic structure of Deep Convolutional Network (DCNN), a DCNN network model suitable for the research scenario of this paper is established, and the model is utilized for feature extraction of images in multi-view videos.The softargmin algorithm is used to generate the parallax map for parallax estimation in the parallax calculation stage.According to the parallax map, voxel-based 3D reconstruction of the target in the multiview video is performed, and the surface reconstruction of the voxel model is performed using the Marching Cubes algorithm, and after obtaining the surface model of the target object, texture mapping is performed to enhance the realism of the model.The deep convolutional network based 3D building method in this paper can effectively realize the feature extraction of target objects in multi-view video.In 3D target modeling, the model in this paper achieves good results on both public and measured datasets, and has obvious performance superiority and generalization ability compared with other methods.