Human Body Parameter Measurement Based on Neural Networks and Curve Fitting

The measurement of human body parameters has many applications in the fields of garment making, car seat and so on. In order to solve the problems of many fixed conditions and large errors in traditional methods, a set of human body parameter measurement system based on neural network is designed. The system uses OpenPose and Deeplabv3+ algorithms to detect human feature points and extract contors respectively, and calculates and fits parameters by B-spline curve, Euclidean and other methods, and then uses deep neural network to accurately convert the pixel distance of human parameters into the real distance. The results show that the average error of 8 human parameters is 3.24%, and the measurement time is less than 15s. The results are consistent with the actual measurement requirements and have certain robustness.