Clothing Extremity Identification Using Convolutional Neural Network Regressor

Understanding and manipulating a textile objects with a high-dimensional configuration space in relation to its context poses a considerable challenge in the area of Robotics. One of the first step for manipulating textiles is to identify key grasping points on extremities such as collars and hem in order to have a context-aware robotic grasping system. In this study, we proposed a method for identifying clothing extremity using a Convolutional Neural Network as a bounding box regression approach. Results indicate that the said method was able to identify and discriminate features of the collar while providing a high accuracy on identifying collar keypoints through a bounding box approach.