Checkerboard Projective Human Body Size Measurement System Based On CAFA-PSPNet

Human body size data plays a vital role in human health assessment,private customization,and the development of the clothing industry.With people’s emphasis on privacy,there is an urgent need for a safe,convenient human body size measurement method with high-precision.Traditional body size measurement technologies cannot meet the current demand,and noncontact body size measurement has become a hotspot.At present,for most non-contact human body measurement methods,the subject is generally required to wear special clothes,and the operation process is complicated.To solve the above problems,a checkerboard projective human body size measurement system based on CAFA-PSPNet human body segmentation network is proposed in this paper.The detailed research contents are as follows:1.A CAFA-PSPNet human body semantic segmentation algorithm is proposed in this paper to further improve the segmentation accuracy of the network model.The Coordinate Attention(CA)mechanism module is added to improve the feature learning and generalization ability of the network model.Meanwhile,the Feature Aggregation(FA)module is integrated to reduce the complexity of the model.In the comparison experiments of segmentation networks,the proposed CAFA-PSPNet segmentation network showed good performance in segmenting human body images captured by the binocular camera.When compared with PSPNet,MIOU,MPA and PA increased by 1.74%,0.77% and 0.13% respectively,while the parameter amount and FLOPs are slightly increased.Moreover,when compared with CA-PSPNet which only adopts CA improvement and has comparable accuracy performance,the parameter amount and FLOPs are reduced by 2.49 M and 10.28 G.2.A Prewitt-Hough circle detection algorithm is designed to use in the key point recognition.The algorithm solves the problem of color dependence in HSV color space-based key point detection algorithms through Hough circle detection.The edge information extraction ability for key points is increased by introducing Prewitt operator which solves the problems of parameter selection sensitivity,leak detection and noise sensitivity in the Hough circle detection algorithm.Experimental results show that the algorithm has higher detection accuracy and better anti-noise stability.3.A non-contact human body size measurement system that projects a checkerboard pattern to the subject by a projector equipment is designed to solve the problem that the subject needs to wear specific clothes and has limited freedom when measuring.During the shooting process,the projected checkerboard is used to enhance the corner point features of the subject,which effectively solves the problem of low detection accuracy for feature points.Experimental results show that the average absolute errors of bust,waist and hip measured by the proposed system are 0.640 cm,0.761 cm and 0.676 cm respectively,which can provide reliable measurement data for the subjects.The checkerboard projective body size measurement system based on CAFA-PSPNet human body segmentation network proposed in this paper has high measurement accuracy,and may promote the evolution of non-contact body size measurement.