TY - JOUR T1 - Classification and Identification Model of Young Women's Torso Shape Based on Human Surface Curve Features AU - Pan , Rou-Xi AU - Yu , Chen AU - Guo , Rui-Liang JO - Journal of Fiber Bioengineering and Informatics VL - 1 SP - 69 EP - 87 PY - 2023 DA - 2023/10 SN - 16 DO - http://doi.org/10.3993/jfbim02172 UR - https://global-sci.org/intro/article_detail/jfbi/22061.html KW - Torso Shape KW - Body Angles KW - Body Shape Classification KW - Deep Learning AB -

Human body shape analysis is an important reference basis for garment sizing and modification. The study of human body shape is to better master the relationship between the size and shape of different body parts and the overall shape of the garment. In this paper, 245 young women aged between 18 and 24 years in school in northern China were selected as the study subjects by applying the 3D human body measurement technology. Using the statistical software SPSS, principal component analysis, correlation analysis and R-type clustering were performed to evaluate 16 variables, including height, girth, and body surface angles. Five body angles were extracted as classification indexes: chest angle, back inclination, dorsal angle, body lateral angle, and buttocks angle. These indexes were critical in explaining the characteristics of the torso surface curve. Consequently, the body types were divided into three categories using K-means clustering. More detailed characteristics of the eight body types ${\rm Y}_{{\rm II}}$ to ${\rm B}_{{\rm III}}$ were classified by combining the chest-waist drop of the Chinese National Standard classification indication. Then according to the classification results, a recognition template that can automatically classify body shape was created through the Baidu AI EasyDL development platform. Experimental results showed that the average precision of the body type recognition model reached 91.7%, among which the recognition accuracy for Type III S body shape was over 95%, providing a meaningful reference for body type classification research.