Research On Real-time Personalized Human Body Model Reconstruction Algorithm Based On Two-dimensional Image

With the development of computer and Internet technology in the apparel industry,image-based non-contact anthropometric measurement has gradually become the research trend of human body measurement technology with its advantages of fast and accurate extraction of human body size data,and is widely used in clothing customization,human body modeling,virtual fittings and other fields.The accuracy of human body size measurement depends on the accuracy of human contour extraction,feature point extraction,and circumference fitting.In this thesis,the existing anthropometric algorithms are improved on the problem that the contour and feature points are not accurately extracted.This thesis also summarizes the advantages and disadvantages of the existing perimeter fitting method.Finally,the human body dimension measurement method based on orthogonal image of human body is proposed,and the automatic human body model automatic generation system is constructed.The main research contents are as follows:(1)Aiming at the problem that the traditional contour extraction method is not accurate in complex background,an improved adaptive human contour extraction method is proposed.The skin color area and the clothing area are used as the target areas.The Gaussian model is used to extract the skin color region.The H component in the HSV color space has characteristics independent of light changes and is used to extract the clothing area.Finally,the two regions are linearly fused to achieve the extraction of human contours.The method effectively reduces the limitation of conditions such as complex background environment,and can obtain a relatively accurate and complete human body contour.(2)A human body size measurement algorithm based on human body feature region localization and pixel scanning is proposed,and a parametric model is constructed.First,the scanned image extracts the top coordinates and the sole coordinates to obtain the pixel height of the human body in the image.Then,the approximate feature area of the remaining feature points is inferred in combination with the human proportional relationship,and local pixel scanning is performed in this area to extract the remaining feature points.Then,the actual size of the human body is calculated by size calibration and fitting.Finally,the standard 3D human body model is used as the basic model to establish the rules of variation between the human body size data and each coordinate to realize the parametric model.(3)In order to meet the individual needs of users,an image-based personalized human body model automatic generation system is designed and implemented in this thesis.The system automatically measures the user's size based on the input user images and uses the data to drive the standard human body model deformation to generate a human body model similar to the user's body.This system uses Unity3 D as the development platform and C# as the writing language.It can read the positive and side images of the human body by uploading photos or taking photos in real time,and extract feature points and feature sizes,and display the personalized human body model.In the thesis,there are 38 pictures,10 tables and 63 references.