| 摘 要: 针对研究神经辐射场重建三维网格,从多视图照片中提取出高质量的网格模型,文章提出了一种基于NeuS的新型网络模型结构Texture Enhancement NeuS(TE-NeuS),该模型结构具备纹理增强特性,通过分解采样光线,运用多层感知机将一条模糊光线建模成5条清晰光线的加权,能够从轻微晃动的人体照片中提取出高质量的人体头部三维网格模型。用5组人头照片进行测试,实验结果表明,相较于NeRF和NeuS,TE-NeuS在倒角距离指标上分别提升了21.89%与13.36%,并且成功地提取出高质量的人体头部三维模型。 |
| 关键词: 三维重建;数字人;神经辐射场;神经表面重建 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 国家自然科学基金面上项目(61672466);国家自然科学基金委国际合作与交流项目(62011530130) |
|
| 3D Reconstruction of Human Head Based on TE-NeuS |
|
ZHANG Jingwen1, ZHANG Haixiang1, LI Shaohua2, JIANG Mingfeng1
|
(1.School of Computer Science and Technology, Zhejiang Sci-Tech University, Hangzhou 310018, China;
2.School of In f ormation Science and Engineering, Zhejiang Sci-Tech University, Hangzhou 310018, China)
22944746@qq.com; 524126687@qq.com; 1293694896@qq.com; m.jiang@zstu.edu.cn
|
| Abstract: In order to reconstruct a 3D mesh of the neural radiance field and extract high-quality mesh models from multi-view photos, this paper proposes a novel network model structure called Texture Enhancement NeuS (TENeuS) based on NeuS. The proposed model structure has texture enhancement characteristics and can decompose and sample rays. It utilizes a multi-layer perceptron to model one fuzzy ray into five clear rays with weighted values. This enables the extraction of high-quality 3D mesh models of the human head from slightly moving body photos. The model is tested using five sets of head photos, and experimental results show that compared to NeRF and NeuS, TENeuS improves the chamfer distance indicator by 21.89% and 13. 36% , respectively, successfully extracting highquality 3d models of human heads. |
| Keywords: 3D reconstruction; digital human; nerve radiation field; neural surface reconstruction |
