APP下载

多视角单目服装图像三维重建算法的研究与实现

2021-11-26胡新荣曾潇刘军平彭涛何儒汉陈常念

现代纺织技术 2021年6期

胡新荣 曾潇 刘军平 彭涛 何儒汉 陈常念

摘 要:为降低服装三维模型的重建难度,提升三维服装模型的重建效率,提出一种基于多视角单目的服装三维重建方法,该方法首先获取服装图像序列,然后对服装图像序列进行实例化分割获得含服装部分的轮廓信息,使用SIFT算法提取每张图像的特征点和匹配对,加入双重约束去除误差匹配对。然后进行稀疏点云以及稠密点云的重建,最后使用泊松重建还原服装表面细节。结果表明:在多视觉单目服装三维重建过程加入实例分割和双重约束后可以将特征的匹配率提升至88.3%,三维重建的速度较传统方法提升了22%。该方法在服装三维模型重建的过程中也能较好还原服装表面细节。

关键词:服装重建;实例分割;特征点检测;特征点匹配;点云

中图分类号: TP391.41

文献标志码:A

文章编号:1009-265X(2021)06-0084-07

Research and Implementation of 3D Reconstruction Algorithm for Images of

Multi-view Monocular Garment

HU Xinrong, ZENG xiao, LIU Junping, PENG Tao, HE Ruhan, CHEN Changnian

(School of Computer Science and Artificial Intelligence, Wuhan Textile University, Wuhan 430200, China)

Abstract: In order to reduce the difficulty of 3D garment model reconstruction and improve the efficiency of 3D garment model reconstruction, a 3D reconstruction method based on multi-view monocular garment is presented. In this method, the garment image sequence is firstly obtained, and then the contour information of garment is obtained by instantiating and segmenting the garment image sequence. The feature points and matching pairs of each image are extracted by using SIFT algorithm and error matching pairs are eliminated by adding dual constraints. Then, sparse point cloud and dense point cloud are reconstructed, and finally the surface details of the garment are restored by Poisson reconstruction. The results show that after adding instance segmentation and dual constraints to the 3D reconstruction process of multi-view monocular garment, the matching rate of features can increase to 88.3%, and the speed of 3D reconstruction can increase by 22% compared with the traditional method. This method can also restore the surface details of the garment properly in the process of 3D model reconstruction.

Key words: garment reconstruction; instance segmentation; feature point detection; feature point matching; point cloud

收稿日期:2020-03-20  網络首发日期:2020-12-17

基金项目:国家自然科学基金项目(61103085);湖北省高校优秀中青年科技创新团队计划项目(T201807)

作者简介:胡新荣(1974- ),女,武汉人,教授,博士,主要从事图形图像处理方面的研究。

随着互联网技术和虚拟现实技术的发展,三维服装模型广泛应用于服装设计、个性化定制、虚拟试衣、影视娱乐、动漫游戏各领域[1]。三维服装模型的制作同时也吸引了越来越多服装及图形学相关领域研究者的兴趣。由于服装具有材质柔软、易变形、易褶皱及结构复杂等特性,使得服装三维建模相较于其他刚性物体建模具有更大的难度。……

登录APP查看全文