Predicting human dimensions in body images for automatic generation of female pants patterns

Abstract

This study proposed a method on how to obtain and predict body measurements from frontal and side images of a subject for the individualized pattern generation of women's pants. According to the relationship between front and back patterns and a body torso, 32 important pattern dimensions relevant to certain body dimensions were determined by the graphic flattening method. For the body dimensions (such as perimeters) that could not be directly extracted from the body images, the prediction models were established based on the available width and depth measurements. The body measurements from the body images of 425 subjects were compared with the corresponding manual measurements, which showed a good correlation between the automated and manual measurements. The tried-on test showed that the pants made with the generated patterns demonstrated good fitting effect at the important characteristic landmarks of a participating subject. This method can accelerate the pattern-making process for women's pants based on body measurements, reducing human efforts, costs and production time.