Robust Color Object Recognition for a Service Robotic Task in the System FRIEND II

One of the key requirements of service and rehabilitation robotic systems is the robust perception of the environment. This paper presents the machine vision framework within the rehabilitation robotic system FRIEND II and a method for improving the visual perceptual capability of this system. For the reliable, autonomous performing of the "beverage serving" task robust visual information about the objects to be manipulated in the presence of variable lighting conditions is necessary. A novel aspect of the proposed color object recognition method is its use of feedback control at the image segmentation level, which makes it able to cope with the object color uncertainty caused by different illumination conditions. The idea behind the closed-loop image segmentation is to provide the object recognition step with input data on which it can rely. The proposed object recognition method uses Hu moments which are invariant to rotation, translation and scaling of the object in the image. The benefit of the closed-loop color object recognition in service robotics is demonstrated through the example of the recognition of the green bottle. The presented experimental results on the performance evaluation show that the proposed method has robustness with respect to illumination as well as with respect to the different localization of the object to be manipulated.