The major goal of this project is to develop an automated apparel manufacturing system. Given a roll of fabric, the system takes different pieces of fabrics and produces customized jeans or clothes. During the fabrication process, the cut fabrics travel through the work-cell from the loading zone to the transportation zone, integration zone, and finally the sewing zone. Our main objective, as a vision group in this project, is to constantly monitor the fabric in real-time and provide feedback to the control system. We have developed two vision systems across the entire work-cell: overhead vision system (OHVS) and high-speed vision system (HSVS).
The overall objective of HSVS is to track denim fabrics with repetitive weave patterns in a non-Euclidean thread-based coordinate system. In the sewing zone of an automatic sewing machine, in order to provide feedback for the control system, HSVS measures fabric position and orientation in real time and high accuracy. Previous research generally uses the shape information of fabrics to provide position feedback, which is sensitive to noise in the fabric edge and fabric deformations off the table surface. To alleviate this issue, we design algorithms for tracking denim fabrics and counting fabric threads allowing for fabric distortions. One example of texture tracking with thread counting is as below. Given two consecutive denim texture frames with a small motion, HSVS reports the following outputs: (1) a detected lattice model consistent with the underlying topological relationship among repetitive weave patterns (i.e., a thread-based coordinate system): A0 and B0 mean two basis vectors of a local fabric lattice represented by pixels of a conventional coordinate system (i.e., the camera space); (2) translation parameters in a conventional coordinate system: dx and dy represent translation parameters in the horizontal direction and the vertical direction, respectively; (3) translation parameters in the thread-based coordinate system: dA and dB denote translation parameters in A0 and B0; (4) a rotation angle: dtheta is a rotation angle in cartisian degrees.
Y. Hu, Z. Long, and G. AlRegib, “A high-speed, real-time vision system for texture tracking and thread counting,” IEEE Signal Processing Letters, 2018. [PDF]