基于二维图像的三维服装重建  

Image-based three-dimensional garment reconstruction

作　　者：潘博 钟跃崎[1,2] PAN Bo;ZHONG Yueqi(College of Textiles, Donghua University, Shanghai 201620, China;Key Laboratory of Textile Science & Technology, Ministry of Education, Donghua University, Shanghai 201620, China)

机构地区：[1]东华大学纺织学院,上海201620 [2]东华大学纺织面料与技术教育部重点实验室,上海201620

出　　处：《纺织学报》2020年第4期123-128,共6页Journal of Textile Research

基　　金：国家自然科学基金项目(61572124)。

摘　　要：针对泊松重建过程中点云缺失导致曲面错误重构的问题,提出采用最近点迭代技术对分批重建点云实现配准融合,并以此恢复三维结构的解决策略。通过对比不同拍摄方案和图像数量对模型重建效果的影响,确定了合适的重建图像数量与拍摄方案;对泊松重建八叉树深度的最优参数选择进行分析,并在此基础上对重建模型精度进行探究。结果表明:泊松重建八叉树深度为11时,可还原模型表面细节;图像数量大于60,且采用半球式拍摄方案更有利于模型的完整性;以深度相机扫描获取点云作为基准,最终获取的三维模型误差小于5.8 mm。Due to mesh reconstruction error caused by the lack of point cloud during the Poisson reconstruction,a solution strategy that fuse dense point cloud reconstructed in batches by using the nearest iteration algorithm was proposed to restore three-dimensional(3-D)structure.The appropriate number of reconstructed images and shooting schemes were determined by comparing the effect of reconstructed dense point cloud model.The optimal parameter selection of octree depth during Possion reconstruction were analyzed,and the model accuracy based on these strategy was tested.The results indicate that it is sufficient to recover model surface details when octree depth has been set up to 11.The model demonstrates more integrity when the image quantity is greater than 60.Adopting"Hemispherical"shooting scheme has been proven effective in enhancing the model integrity.The error of final 3-D model is less than 5.8 mm by taking the point cloud obtained using depth camera sensor as a benchmark.

关 键 词：泊松重建 二维图像 三维重建 点云融合 最近点迭代技术 八叉树深度 服装虚拟展示 

分 类 号：TS942.8[轻工技术与工程—服装设计与工程]