Augmented Reality Animation Image Information Extraction and Modeling Based on Generative Adversarial Network

Abstract

Modeling is the first step of 3D animation design and production, and it is the core and foundation of the 3D world. Without a good model, other good effects are hard to show. However, the traditional geometric modeling method has complex modeling process and poor realism. In this article, Gan (Generative Adversarial Networks) model in AI (Artificial Intelligence) is applied to the information extraction of AR (Augmented Reality) animation images, and the modeling process of 3D animation CAD is optimized. In the information extraction model of AR 3D animation image based on GAN, a MFFM (Multi-Feature Fusion Module) is introduced into the traditional generator, which combines the characteristics of different expansion convolution rates. Moreover, the loss function of feature reconstruction based on perception is introduced into the generator, which is convenient for the network to extract more abundant features. In order to verify the reliability of this method, the GAN constructed in this article is compared with several other neural network models. Simulation results show that compared with the classic CNN model and the classic BPNN model, the error of the proposed GAN model is the lowest, and finally it is stable at about 0.54. And the accuracy of the network model is excellent, which can basically reach more than 90% and the highest can reach 95.24%. This shows that the research on information extraction and modeling of AR 3D animation images in this article has achieved good results.