Research On Controlling Gesture Recognition Technique And Application In Non-Contact HCI

With the development of computers, traditional human computer interaction (HCI) methods, such as mice, keyboards and so on, are showing their disadvantages in the new application scenarios for their inconvenient and unnatural communicating way. Although touch screen based HCI can offer the users natural, convenient and understandable communication way, and they are already widely used in mobiles and tablet PCs, their application conditions also be limited for their touch interaction requirements. For example, doctors can’t use the touch interaction method when they are performing the operation. In order to overcome the drawbacks of HCI based on touch ways, this thesis focus its attention on non-contact HCI based on Kinect, and get some good results as follows.First, the key point of hand gesture for non-contact HCI is to recognize the users’hand control gestures, this thesis chooses Kinect as input devices and implement machine learning method to recognition hand gestures. Considering the class-imbalance problem, a cascade method named GesCasc based learning algorithm is proposed to figure out the controlling hand gesture recognition problem. From the experiment, the proposed method GesCasc gets good performance.Second, when the non-contact HCI method is mass used, the trained unified recognition model can’t satisfy all users for different users may have different gesture styles. In this thesis, a method named PerGrab is proposed to deal with this problem. When specific user can’t get good recognition result, he or she will be asked to collect some gesture data and label them, and then PerGrab can help the specific user get his own personalized hand gesture recognition model with only little efforts.Third, this thesis designs and implements a non-contact hand gesture HCI application based on Kinect named PerControl. In PerControl, users can use their hand gestures to operate the shown images, they can zoom, move and rotate the images to change the locations and angles of the shown images, and then they can get good views of those images.