基于多视角图像三维重建与点云配准的着装间隙量表征  

Characterization of clothing ease allowance based on the 3D reconstruction and point cloud registration of multi-view images

作　　者：卢志洁 侯珏 杨阳[1,2] 刘正[2,3,4] LU Zhijie;HOU Jue;YANG Yang;LIU Zheng(School of Fashion Design&Engineering,Ministry of Culture and Tourism,Zhejiang Sci-Tech University,Hangzhou 310018,China;Zhejiang Provincial Research Center of Clothing Engineering Technology,Ministry of Culture and Tourism,Zhejiang Sci-Tech University,Hangzhou 310018,China;International Institute of Fashion Technology,Ministry of Culture and Tourism,Zhejiang Sci-Tech University,Hangzhou 310018,China;Key Laboratory of Silk Culture Heritage and Products Design Digital Technology,Ministry of Culture and Tourism,Zhejiang Sci-Tech University,Hangzhou 310018,China;Wuhan Textile and Apparel Digital Engineering Technology Research Center,Wuhan Textile University,Wuhan 430073,China)

机构地区：[1]浙江理工大学,服装学院,杭州310018 [2]浙江理工大学,浙江省服装工程技术研究中心,杭州310018 [3]浙江理工大学,国际时装技术学院,杭州310018 [4]浙江理工大学,丝绸文化传承与产品设计数字化技术文化和旅游部重点实验室,杭州310018 [5]武汉纺织大学武汉纺织服装数字化工程技术研究中心,武汉430073

出　　处：《浙江理工大学学报（自然科学版）》2024年第2期202-209,共8页Journal of Zhejiang Sci-Tech University(Natural Sciences)

基　　金：浙江省科技计划项目(2023C03181)。

摘　　要：为了分析人体不同姿势下的着装间隙量,采用多视角图像三维重建方法获取净体和着装的三维点云,利用OpenPose人体姿态估计模型确定关节点并逆投影到三维空间,获得净体和着装两层点云粗配准的匹配点坐标;基于着装未覆盖的人体点云,通过最近点迭代(Iterative closest point,ICP)算法实现两层点云的精配准,并对配准模型进行验证,建立三维着装模型。根据三维着装模型计算两层点云间的最近邻域距离表征着装间隙量。结果显示:经两步配准后,躯干着装未覆盖处两层点云距离平均值和均方根误差分别为2.859 mm和3.260 mm,胳膊区域两层点云距离最大的平均值和均方根误差分别为4.018 mm和4.735 mm,配准结果精准度较好。该方法可以准确地获取不同姿势的着装间隙量的大小与分布,为分析着装间隙量与运动的关系、评价舒适性、建立热传递模型及评估虚拟试衣效果提供关键技术手段。To analyze the amount of clothing ease allowance under different postures of the human body,the three-dimensional(3D)point cloud of the nude body and clothing was obtained by 3D reconstruction based on multi-view images.Using the human pose estimation model of OpenPose,we estimated the joints and reversely projected them to the 3D space to get the coordinates of the matching points for the coarse registration of the two layers of point clouds of the nude body and clothing.We realized the accurate registration of the two layers of point clouds based on the uncovered human body point clouds by the iterative closest point(ICP)algorithm,and verified the registration model,building a 3D dressing model.Based on the model,we computed the nearest-neighbor distance between the two layers of point clouds,and defined the distance as the ease allowance.The results showed that after twostep registration,the average distance and root mean square error of the two layers of point clouds in the uncovered area of the torso were 2.859 mm and 3.260 mm,respectively,and the maximum average distance and root mean square error of the two layers of point clouds in the arm area were 4.018 mm and 4.735 mm,respectively,indicating good accuracy in registration results.This method can accurately obtain the size and distribution of clothing ease allowance for different postures and provide key technical means for analyzing the relationship between clothing ease allowance and movement,evaluating comfort,establishing heat transfer models,and evaluating virtual fitting effects.

关 键 词：着装间隙量 三维重建 点云配准 三维着装模型 多视角图像 

分 类 号：TS941.2[轻工技术与工程—服装设计与工程]