基于空间向量模长的青年女性腰腹臀部形态分类与判别

Classification and discrimination of waist-abdomen-hip morphology of young women based on space vector length

为实现女体三维曲面的分类与判别,提高裤装的合体性,提出一种基于空间向量模长表征腰腹臀部形态并进行体型分类的方法。获取323名18~25岁青年女性的三维点云数据,提取腰腹臀部10层横截面曲线,以臀围质心为原点重建点云坐标系,通过欧式距离计算130个特征点的空间向量模长,构建表征人体曲面形态的模长矩阵。引入拉普拉斯特征映射降维获取18个本征维度,采用K-means聚类,运用随机森林算法建立青年女性腰腹臀部形态的判别模型。结果表明,青年女性腰腹臀部可分为大众型、扁平型、腹凸型、臀凸型4类,分别占样本总数的58.82%、27.86%、8.36%和4.95%。分析获得4类体型对应的裤装样板差异,有效提高了裤装合体性,青年女性腰腹臀部形态判别准确率达96.92%。

Objective The current research on body shape uses classification indexes for body shape such as circumference,width,thickness,ratio and angle,which are not able to fully reflect the curved shape of body.In order to establish classification and discrimination of three-dimensional surfaces of female body and improve the fit of pants,a method of body shape classification based on space vector length characterization of waist-abdomen-hip morphology is proposed.Method The 3-D point cloud data of 323 young women aged 18-25 years were collected by TC 23D scanner,and the 10-layer cross-sectional curves of the waist-abdomen-hip were extracted.The center of mass of hip circumference was used as the origin to reconstruct the point cloud coordinate system.The space vector length of 130 feature points was calculated by Euclidean distance to construct the modulus length matrix to characterize the morphology of body surface.The eigen dimension was determined by using maximum likelihood estimation,and Laplace feature mapping was introduced.The body type segmentation based on space vector length was achieved by K-means clustering.The discriminative model of waist-abdomen-hip morphology of young women was established by random forest algorithm.The garment samples(SVP)corresponding to four types under size 160/66A were obtained and compared with the basic women's suit pants benchmark sample(BTP)to evaluate their fit.Results Based on the 3-D point cloud data,10 feature cross-sections reflecting the morphology of the waist-abdomen-hip surfaces were extracted,and 130 feature points were extracted.A modal length matrix of 323 samples,each with 130 space vector lengths was constructed.The maximum likelihood estimation was used to determine the eigen dimension as 18.The modal length matrix 32318 characterizing the body surface morphology was obtained by dimensionality reduction.The elbow method was used to determine the number of clusters as 4.Through cluster analysis,the waist-abdomen-hip of young women were subdivided into four types by combining the national standard body type,i.e.,mass,flat,convex abdomen and convex hip,which accounted for 58.82%,27.86%,8.36% and 4.95% of the total number of samples respectively.In order to perform the discrimination test for the independence of different body types,a discriminative model of waist-abdomen-hip morphology of young women was established with a decision tree number of 100,a classification number of 4,a training and testing sample of 8∶2.The model was continuously trained by Random Forest(RF)algorithm,and the discrimination accuracy reached 96.92%.Using the virtual fitting,it can be seen that the SVP sample pants are blue and green areas at the waist and crotch.The garment stretch is between 100%-110% in a normal stretch state,with moderate dressing pressure and a good garment fit.Through virtual fitting and physical verification,the fit of SVP sample pants became better than that of BTP sample pants.Conclusion The space vector length is used as a classification index to characterize the surface morphology of the waist-abdomen-hip of young women and to perform body type segmentation to achieve the classification and discrimination of female body 3-D surface.The discrimination test of the independence of different body types shows that the four types classified by the space vector length as the index show strong independence.Through virtual validation and physical validation,it is shown that the body type classification method based on space vector length has a good effect on improving the fit of pants.The application of this method can effectively facilitate the body type segmentation and further improve the matching degree of clothing and human body,which can be applied to the scale customization of garment.