Body size measurement based on deep learning for image segmentation by binocular stereovision system

This paper proposes a body size measurement system based on deep learning for image segmentation by binocular stereovision for clothing design. A pair of binocular stereo cameras and a precise revolving platform are used to collect the stereo images of the entire human body. A more diverse human body dataset is constructed for deep learning. An optimized girth semantic segmentation model based on PSPNet network is well trained by deep learning to realize the automatic segmentation of more girth regions in human body images taken from multiple angles. Color subspaces with larger spacing in color spaces are designed to perform more effective stereo matching. The space coordinates of 3D point corresponding to each stereo matching point pair are calculated in the same coordinate system. The space coordinates are reversely revolved back to the intial coordinates of the same coordinate system according to the revolving angle of the platform. Girths of bust, waist, hip and thigh are measured by calculating the fitting curve length after girth fitting of the markers thereon. Height of bust, waist, hip, thigh and entire body are estimated according to the ratio of the 3D distance to the number of pixels between two adjacent markers on the same girth. Quantitative experiments on many human subjects demonstrate that the proposed system can measure body size intelligently and accurately for clothing design under the regulation of China national standards GB/T 2664-2017, GA258-2009, GB/T 2665-2017 and textile industry standard FZ/T 73029-2019, FZ/T 73017-2014, FZ/ T 73059-2017, FZ/T 73022-2019.