Research Of Young Female Body Shape Recognition Based On Longitudinal Section Curve Form

With the advent of3D non-contact body measurement technology, the human bodydata acquisition becomes faster and more comprehensive; it also provides the conditionfor the study of the human body. But the original point cloud data has a large quantity,less useful information, cannot be easily used by the apparel industry. The currentnational standard divides the human shapes according to the chest-to-waist difference; itis more effective for lateral body shape, which is lack of the distinction between thelongitudinal and cross section shape characteristics. Moreover, there is little research onlongitudinal section curves which are closely connected with clothing structure and cancharacterize the human body size. In addition, the clothing structure design onlyconsidered surrounded degree and ratio, ignored the surface curve shape, which reallyaffected the fit of clothing. Therefore, dealing with three dimensional human point cloudto study the longitudinal and cross section shape has theoretical and practical value.This paper studied the human body based on the longitudinal and cross section curve,used reverse engineering technology to reduce dimensions of point cloud data, built themathematical model of the longitudinal and cross section curve, quantified the curveshape of longitudinal and cross section which can represent the human body, it was asupplement to the existing research of human body, providing the basis for the line,provincial road location and size of the clothing structure design. Constructing theprobabilistic neural network model at the same time, and developing a shape recognitionsystem to lay a solid technical foundation for personalized clothing customization,computer aided design and virtual fitting.This paper mainly includes the following aspects of content:1. This paper collected both body dimension data and3D point cloud data. Bychoosing childless female college students aged between18to24in school, used theAmerican [TC]23D human body measurement instrument to measure631samples.2. This paper extracted the characteristic curves of longitudinal section which can beused for shape classification. Using the Imageware12.0to extract three-dimensional pointcloud data of the middle sagittal plane、coronal plane，the sagittal plane across bust point、the sagittal plane across the most salient point of back and the sagittal plane acrossshoulder point, used the least squares method of data fitting, built the high goodnessfitting model of the two-dimensional curve to represent the three dimensional feature space of the body effectively.3. This paper established a method of young female body classification which basedon the radius of curvature of the feature points of the longitudinal and cross section curve.Based on the national standard，subdividing young female body type by analysis the curveshape on the longitudinal and cross section, and pseudo F statistic as a discriminatefunction to determine the optimal class number. By means of analyzing the convexcurvature radius based on the feature points of longitudinal and cross section curve: sideneck point, acromion point, bust point, back convex point, side waist point, abdominalconvex point and hip bumps point, to generate the joint characterization of the bodysurface characteristics of longitudinal and cross section, which include the neck, shoulder,chest, back, waist, abdomen and buttocks. It applied K-means clustering algorithm fordynamic clustering, ultimately the longitudinal and cross section shape could be dividedinto eight categories, which presented a new body shape identification:"national standardsize+longitudinal and cross section shape", and combined horizontal and vertical indexesto distinguish the human body shapes preferably.4. This paper constructed automatic recognition of human body based on theprobabilistic neural network model. On the area of human body identification, to enhancerecognition accuracy, applying the probabilistic neural network method to human bodyrecognition research, using Matlab R2012b software to build probabilistic neural networkrecognition model, recognition accuracy up to98.67%.5. This paper used the Qt software to develop young female longitudinal and crosssection shape recognition system V1.0, consisting of four function modules: data entry ofhuman body size, computing feature point of curvature radius, identification of humanbody, displaying longitudinal and cross section curve.