Research Of Somatotype Based On The Sagittal Plane Silhouette Of Human Body

The classification based on the human body shapes and features is very significant for the applications of textile and garment industry.The result of the classification could be employed to guide the produce of clothing and to optimize the design of clothing pattern,which can also lead to the final clothing goods with more comfortable and fitter.Thus,the method,based on body shapes and features,proposed in this thesis is of researchful and industrial importance.First of all,the sagittal plane silhouette that goes through the center-line of human body was extracted from an encapsulated model that was encapsulated by the point cloud dataset captured through range scanning.In order to decrease,even eliminate,the noise accompany with the extracted silhouette,Ellipse Fourier Fitting(EFT)was induced into this method to complete the job.Whereafter,a turning function was used to parameterize the fitted contour and a new turning function curve,which is a critical operator in the following steps,was obtained during this procedure.Then,by analyzing the turning function curve,some features that are capable of demonstrating the difference among various persons are extracted.Finally,K-Means clustering algorithm was adopted to classify the human body shapes based on the sagittal plane silhouettes section and the result of classification was validated by correlational analysis silhouette subsequently.This thesis mainly includes the following aspects:(1)Extracting of sagittal plane silhouette.In order to capture the sagittal plane silhouette,a stereo scanner was used to approach the point cloud data of target shape which was imported into Geomagic studio,a reverse engineering software,to procure the sagittal plane silhouette later.(2)Fitting of sagittal plane silhouette.The silhouette extracted above is a closed curve but is accompanied with some noise.Therefor Elliptical Fourier algorithm which suits for closed curves can be used to the original silhouette such that describe or even eliminate the noisy but persist the nature feature simultaneously.During this step,the optimal harmonics which was 15 and decided by implementing the error analysis over different fitting effects of various harmonics.(3)Feature analysis of sagittal plane silhouette.The parameterization of fitted silhouette was achieved by the turning function,which leads to a new turning function curve as a result.Then five features that can illustrate these curves.Next,Principle Component Analysis(PCA)algorithm was adopt to these five features for the sake of the enhancement of the pertinence between the features for the sake of the enhancement of the pertinence between and the first 3 components with the greatest contributions were chosen as independent features.(4)Classification of body shapes based on the sagittal plane silhouette.According to the result of PCA over sagittal plane silhouette,the first 3 components were chosen as classification indexes for K-Means clustering analysis,and the value of K was determined by Pseudo F statistical function.The result of classification was validated by the correlation analysis which demonstrates that the average correlation coefficient in cluster was higher than 0.9,which belongs to high-correlation relation and is much higher than the average correlation coefficient on all samples.As a consequence,the classification algorithm proposed in this thesis is absolutely effective and it is capable for human body classification.