Body Size Measurement Based On Binocular Stereovision

With the rapid economic development,people’s living standards have greatly improved.Mass-produced clothes can no longer meet people’s needs,so clothing customization has become more and more common.Accurate body size is a prerequisite for clothing customization.Currently,Anthropometric methods include contact and non-contact.The contact measurement process is cumbersome and the accuracy depends on the tailor’s experience.However,non-contact active measurement is usually expensive,bulky,and radiant.The multi-view anthropometric system proposed in the early stage of our lab used six identical industrial cameras to simultaneously collect stereo images of the human,combined with neural network,image stereo matching,spatial coordinate conversion and curve fitting to achieve girth measurement.The system realizes the passive and intelligent measurement of bust and waist in clothing design,and overcomes the problem of too many invalid matching points in girth measurement.However,it still has some shortcomings,such as few measurement parts,and the inability to realize automatic segmentation of the measurement part region of the flipped human image.On the basis of the existing multi-view system,the precision turntable is added,the system algorithm is continuously optimized and improved,and a human body size measurement system based on binocular stereovision is proposed.The research content of this article includes the following three parts:1.We use the binocular stereovision system builted to collect the whole-person full-girth stereo image of the subject.By enhancing the diversification of the dataset,the semantic segmentation model based on GSS-PSPNet is used to realize the automatic segmentation of the head,bust,waist,hip,thigh,and foot of the flipped human image.2.By optimizing the design of markers for measuring girth,a marker type with a higher detection rate is selected.The automatically segmented girth image is matched to obtain the pixel coordinates of markers,and then the three-dimensional coordinate values of markers are obtained by the principle of binocular three-dimensional measurement.3.The curve rotation principle is used to obtain the initial coordinates of the girth markers,and these markers are curved to obtain a fitting curve,and finally the curve length obtained as the measured girth size.The two-view body size measurement system proposed in this paper solves the problem that the original multi-view anthropometric system cannot realize the automatic segmentation of multiple parts of the flipped human image.At the same time,it realizes the accurate measurement of bust girth,waist circumference,hip circumference,thigh circumference,height,bust height,waist height,hip height,and thigh height.By measuring subjects of different body types and different genders,the absolute error range of girth size is [0.51,1.89]cm,the maximum relative error rate is 3.66%,and the absolute error range of height size is [0.31,3.50]cm,the maximum relative error rate is 3.09%.The measurement errors of bust,waist,hip,and thigh all comply with the corresponding national standards GA 258-2009,GB/T 2665-2017,GB/T 2664-2017 and textile standards FZ/T 73029-2019,FZ/T 73017-2014,FZ/T 73059-2017,FZ/T 73022-2019.Experiments prove that the anthropometric measurement system proposed in this paper has good reliability and accuracy.