Verification of illumination tolerance for photo-model-based cloth recognition

摘要

AbstractGlobally, robots are being used in garment factories according to their advantages such as high precision, flexibility, and productivity. However, it is difficult for robots to deal with deformable objects such as cloths and strings automatically. The main tasks of such robots are to recognize a deformable object and to pick up and place it at a designated position automatically. The deformable character of cloths seems a main hindrance for automatic handling by robots, especially if the cloth is unique. In this paper, a cloth handling robotic system for unique cloths is proposed with consideration of verification of illumination tolerance. This robotic system comprises two main portions: the first portion generates a model of cloths using an image taken by a single camera, and the second portion estimates the relative pose of cloth appeared in the view of two cameras using the generated model. In our cloths’ pose estimation, the photo-model projected from 3D to 2D is used, where this system does not need defining the object’s size, shape, design, color, and weight. The illumination tolerance of the proposed system under different light conditions of different light sources was verified experimentally for evaluating the proposed system from the view point of practicality. The fluorescent light and the light-emitting diode (LED) light are used in this experiment, having confirmed that the proposed system can recognize cloths in condition that the light environments have varieties.