Cnn And Pca Based Visual System Of A Wheelchair Manipulator Robot For Automatic Drinking

Zhijun Zhang,Shi Mao,Ke Chen,Lin Xiao,Bolin Liao,Chunquan Li,Penzchao Zhang

DOI: https://doi.org/10.1109/ROBIO.2018.8665131

2018-01-01

Abstract:To help the disabled who suffers from paralysis with activities of daily living (ADL) and extend their athletic ability, we propose a shared control wheelchair manipulator robot and fully discuss its visual system. In this paper, we design an assistive robot to automatically detect and perform the drinking task for the disabled, which is one of the basic activities people would perform every day. This system incorporates a visual system to automatically detect and recognize target cups, estimate its orientation and detect the user's mouth location for the manipulator to grasp and deliver. Specifically, the visual system separates the potential objects from the background by applying region growing twice, recognize the target by convolutional neural network and estimate the orientation of target object by PCA based method. Finally, we carried out two experiments to evaluate the effectiveness of this system. The result shows that we achieve an average of 90.3% success rate when detecting and estimating the orientation of handleless cups, and our approach is practical in daily activities.