Feature extraction of garments based on Gaussian mixture for autonomous robotic manipulation

Recent advancements in the domain of robotics have offered support to humans in their everyday activities with the aim of offloading workers from performing repetitive tasks. The present work highlights one such task in garment industry where the robot may find potential to manipulate different garments including developing a number of robotic skills like laundry pile sorting, garment stacking and garment folding/ unfolding. This paper is aimed to study the integration of hardware and software developed for ClopeMa Project on a human-friendly robotic platform, i.e. a Baxter robot that can safely operate side by side with humans. In particular, the paper discusses integration of RGB-D sensor with the ROS environment and studies utility of garment manipulation. The goal is to present a working platform which can autonomously recognize the configuration of a piece of garment spread out on a flat surface. The algorithm for recognizing the garment consists of first applying Gaussian mixture model (MoG) for background subtraction and then using polygonal approximation to acquire feature points for the foreground of the garment. The proposed algorithm is tested online through series of experiments on towel, pants and t-shirts of various colors and materials. Results under varying lightening conditions witness robustness of the proposed scheme.