A Study Of Body Shape Measurement Based On Stereo Vision

In recent years,with the rise of Internet technology,people's consumption outlook is gradually changing,especially for wearing clothes,pursuing a quality of life of comfort,experience,and fashion.To meet these requirements,clothing should start from the root cause and quickly measure the accurate three-dimensional size of the human body.The current three-dimensional size data of the human body can be used for virtual fitting,smart medical care,and clothing customization.This paper mainly consists of real-time 3D reconstruction of multiple 2D images.The reconstructed point cloud model is processed by hole repair,noise reduction,and marginalization.At the same time,based on the point cloud library(PCL)library area frame selection,the 3D coordinates of the point cloud are obtained,and the characteristics are The point is the point of interest diffusion to filter the data,use Matlab to draw the scattered point cloud on the acquired data,and finally get the cross-sectional view of the body part,project the three-dimensional point cloud image onto the two-dimensional plane,and calculate the length of the curve through the integration method.The main contents of this paper are as follows:(1)Sample collection and classification processingThis article uses a camera and auxiliary equipment tripod to first obtain images and size measurements of human bodies of different ages and body types,and collect a total of 60 samples.Take the human body as the center and surround the human body to realize the collection of multi-view images,and filter out clear and feature-not-lost pictures.The size is measured using a standard soft ruler to measure 8parts of the human body,and the measured data is statistically analyzed.deal with.(2)Establishment and processing of human body modelFor each object to be measured,no less than 20 images are selected,and the images are input into the algorithm sift for feature point extraction and matching.The key is to find the positional relationship of the same spatial point in multiple perspectives,and use the descriptor to extract the image.After the feature,the feature points are matched in the remaining images to obtain a sparse point cloud model.Finally,the dense reconstruction algorithm CMVS/PMVS is used to encrypt the image.The point cloud model obtained is part of the point cloud lost to form a hole,and at the same time nearby There is a lot of interference,which will cause a lot of interference in the later point cloud area selection,so it is necessary to repair the model holes,reduce noise and edge.(3)Feature point markingThe acquired 3D point cloud model is composed of countless points with spatial information.Taking the measurement and reconstruction of an object in the sample as an example,the number of points with spatial information acquired is 1996142.For the selection of feature points,refer to "Use Choose from the manual of"Anthropometric Basic Projects for Technical Design",pick up the perfect three-dimensional model with a single point and select the area,and select accurate spatial data points for marking based on the parts defined in the manual.There are 12,and the measurement of each part of the body is realized in turn.(4)Frame selection and data filtering of the measurement areaThe point cloud processing in this study is based on the PCL library of C++.First,the PCL library and VS2013 are configured to realize the environment,and then the perfect point cloud model is imported,and the feature points that need to be marked by the frame selection are found,and the ellipse wrapping method is proposed.Obtain the spatial point data of the part to be measured with the rectangular wrapping method,filter the data according to the range determined by the feature points,and use Matlab to read the filtered data for fitting,compare the polynomial fitting and ellipse approximation methods,and deviate from the measurement part The farther data points are deleted,and finally the ideal curve in this paper is obtained.Finally,verify and analyze the same part of the same measurement object and different parts of different measurement objects,use the measurement method in this article to calculate its length and girth and compare it with the real size and analyze the error.At the same time,compare the application in three-dimensional measurement in recent years.The ellipse approximation method proves the validity of the measurement method in this paper,and the accuracy is high.