Gripping Positions Selection for Unfolding a Rectangular Cloth Product

This paper describes a method of gripping positions selection from an item of rectangular cloth placed on a table. To select two appropriate gripping positions on the cloth with unarranged shape, we propose a novel method. The proposed method is a development of the previous work of the author and uses a convolutional neural network. One characteristics of the method is to directly compute gripping positions coordinates by matrix calculation using feature vectors extracted from the layer at the final stage of the neural network. We introduce an improved mechanism to estimate the position of clothing hem, and also an improvement on the part to calculate the gripping positions coordinates. The effectiveness of the proposed method was verified using images of actual cloth products.