Comparative Analysis of Real-time Surface Image Texture Reconstruction for Avatars in Virtual Scenes Development

Avatars play a crucial role in enhancing immersive experiences in virtual reality (VR) and the metaverse, serving as digital representations of users across various applications, including gaming, education, and healthcare. In educational settings, VR avatars create realistic training environments, such as driving simulators, to improve safety and skill acquisition. In healthcare, avatars assist telemedicine by enhancing patient engagement, particularly for those with visual impairments. Despite their benefits, traditional avatar reconstruction methods are often cumbersome and inefficient, hindering widespread adoption. This study focuses on developing a rapid avatar reconstruction method that combines real-time motion detection with photorealistic surface texture generation. We categorize existing techniques into three main groups: Structure from Motion (SfM), Multi-View Stereo (MVS), and Neural Radiance Fields (NeRF). Each method has its strengths and limitations regarding computational efficiency and quality. Through comparative analysis, this research aims to identify the most effective technique for avatar reconstruction in industrial contexts, ultimately enhancing user experiences in virtual environments and promoting the broader adoption of VR technologies.