Estimation of anatomically salient landmarks for human body shapes

Abstract

Making people’s movements natural is always an important and challenging task in computer graphics. People have different anatomical structures depending on their body shape, which makes all movements different. Therefore, accurate anatomical measurement in digital is an indispensable factor. In this paper, we present a semi-automatic regression method to estimate the location of anatomical salient landmarks for given various human body shapes. Our method first uses the shape descriptor to define the shape of people in the body. And we represent the position of salient landmarks for human body shapes using kernel coordinate system. Then, learns relationship between the shape descriptor and the location of anatomical salient landmarks using Kernel Canonical Correlation Analysis(KCCA). After that, given new human body shape, we can estimate the location of anatomical salient points for that body through learned results.