Research On Human - Computer Interaction Technology Based On NAO Robot

Tracking strategy is crucial when concerned with machine vision,especially in Human-Robot Interaction(HRI).On the purpose of realization HRI with NAO robot, we study thegeneral foreground-segmentation&tracking framework and proposeseveral methods to accomplish the task. We firstly build aforeground-segmentation method with Visual Background Extrator(ViBe)in order to extrack the possible foreground which containing people orinteracting targets so that we could move on to track them.Noticing that the robot itself usually suffers from the lack ofsufficient computation ability to deal with the image stream, which makesit necessary to transport the frame captured by robot to the PC in theimplementation of HRI. In addition, sending commands back to robot istime-consuming, thus we have to develop tracking strategies in theaccount of tracking efficiency. In this paper, two ways and perspectives offast tracking are studied:(1) Based on compressing sensing theory, it is possible to projectfeature vectors of target to a lower space which could cutdown the cost to compute and classify feature vectors.(2) From the view of relation of space, we can evaluate therelationship between targets and background and frame amodel conerning both of them, which gives a brand-new wayto analyse and track targets.