Automated Elbow Point Extraction and Upper Arm Length Estimation from 3D Human Body using Neural Network

Hao-yang XIE, Xi CHEN, Zhi-cai YU, Yue-qi ZHONG

Abstract


Elbow point is a very important landmark for body measurement and certain garments. However, little attention has been invested. It is rather difficult to extract the elbow point based on the commonly used geometric methods since 1) the arm and elbow have high degrees of freedom, and 2) there is no prominent feature when the arm is unbent. In this paper, we use human stature, one of the most accessible measurement, to estimate the elbow point for arbitrary arm poses. Specifically, we first train two end-to-end neural networks for males and females to estimate the Euclidean length of the upper arm, and then propose a framework to approximate the elbow point and to calculate the tape measurement of the upper arm. Experimental results have verified that our method is efficient for clothing applications. The mean squared errors for male network and female network on the test dataset are 3.23 and 2.69, respectively. Both the absolute errors for the tape measurement of the upper arm for males and females are less than 1 cm.

Keywords


3D human measurement, Elbow point extraction, Upper arm length estimation, Neural network


DOI
10.12783/dtcse/msam2020/34263

Full Text:

PDF

Refbacks

  • There are currently no refbacks.