SMPL Variable Model for 3D Reconstruction and Image Fusion in Animation Media Applications

3D human reconstruction is a pivotal research area in computer vision and computer graphics, with broad applications in fields such as virtual reality, animation media, and medical simulation. Traditional reconstruction approaches often rely on fixed-shape models, which limits their ability to accur...

Full description

Saved in:
Bibliographic Details
Main Authors: Tian Xie, Kunpeng Zhao
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10918715/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:3D human reconstruction is a pivotal research area in computer vision and computer graphics, with broad applications in fields such as virtual reality, animation media, and medical simulation. Traditional reconstruction approaches often rely on fixed-shape models, which limits their ability to accurately represent the dynamic soft tissue variations and complex interactions with the surrounding environment. These methods face challenges in managing occlusions in multi-person scenes, capturing a diverse range of poses, and handling extensive deformations in real-world scenarios, which constrains their practical application. To overcome these limitations, this paper introduces an enhanced 3D reconstruction model using the SMPL variable model, which integrates a soft tissue layer, UV space mapping, and a force-conditioning mechanism. The proposed model incorporates a variable-thickness soft tissue layer to better capture anatomical variations, utilizes UV space mapping to streamline deformation calculations, and employs force-conditioning to realistically simulate intricate interactions. Experimental results on datasets such as ShapeNet, ShapeNetCore, ABC, and Scan2CAD demonstrate that our approach surpasses current state-of-the-art methods in terms of accuracy, recall, F1 score, and AUC, while also achieving greater efficiency in both training and inference phases. This research not only delivers a more accurate and computationally efficient solution for 3D reconstruction but also contributes novel insights into multi-person interaction modeling and enhances generalization in real-world animation media applications.
ISSN:2169-3536