A Machine learning approach for generating the crotch curve in customised trouser patterns

Abstract

Pattern making is a crucial process in garment manufacturing, as inaccuracies at this stage can result in excessive garment adjustments, negatively affecting productivity. Pattern makers must possess a high level of skill along with a knowledge of human body anthropometry, fabric properties, and garment fit options. However, small and medium garment industries face challenges due to a shortage of skilled pattern makers. This research shows the inaccuracies inherent in crotch curves derived from customised measurements when traditional pattern cutting methods are applied. To address this issue, this research suggests automated pattern creation methods. Specifically, this study focused on developing solutions for generating crotch curves in trouser patterns, considered one of the most challenging aspects of creating trouser patterns. A machine learning approach was adopted to automate the generation of front and back crotch curves in women’s trouser patterns, using 4 body measurements of women, collected from 512 women aged 18 to 26. A database of measurements was created, and patterns were developed using Seamly2D software based on Aldrich’s pattern making method. A supervised learning approach using a neural network regression model was adopted to predict the front and back crotch curves based on an individual’s body measurements.