Optimizing Augmented Reality Game Rendering Pipeline Using Convolutional Neural Networks: Multi-layer Convolutional Acceleration and Texture Enhancement

Abstract

This paper presents a novel approach to optimizing the augmented reality (AR) game rendering pipeline by leveraging Convolutional Neural Networks (CNNs) for real-time texture enhancement and rendering acceleration. The integration of CNNs into the Unreal Engine rendering pipeline enables the enhancement of low-resolution textures, improving visual quality without compromising performance. The CNN model employed in this study achieves significant improvements in both texture quality, as measured by Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), and rendering speed. The results demonstrate a reduction in rendering time by approximately 25%, while maintaining an average frame rate of 45 FPS, making the system suitable for real-time AR applications. User perception tests confirm that the CNN-enhanced method provides superior visual quality and a more engaging user experience compared to traditional methods. The proposed approach offers a promising solution to the challenges of high-resolution texture rendering and real-time performance in AR games. Future research could explore the use of advanced CNN architectures and hybrid models to further improve rendering efficiency and visual fidelity, particularly for mobile platforms.