Wrinkle Detection and Cloth Flattening through Deep Learning and Image Analysis as Assistive Technologies for Sewing

Abstract

Robotic manipulation of fabric has potential as an enabling accessibility technology for individuals with disabilities, opening up a range of employment opportunities and helping to decrease the underemployment of this population. This research seeks to reliably characterize wrinkles and facilitate robotic removal of the wrinkles, with the focus on managing the outfeed of a sewing process, facilitating employment for individuals unable to reach behind the machine while performing sewing tasks. Outfeed management is critical in sewing to prevent bunching and maintain sewing productivity. To smooth out a fabric and eliminate wrinkles, the wrinkles need to be located and characterized, and points identified where a robotic arm can apply force on the fabric to smooth the fabric. For this purpose, we employ a deep learning technique to detect wrinkles and use corner detection of the fabric to determine an effective point for wrinkle removal.