Wearable Sensor Based Gesture Recognition

With the continus expanding of computer in various areas, keyboard and mouse have become increasingly prominent of its drawbacks and limitations as human interface device. As the promoting of pervasive computing and people-oriented concept, natural HCI (Human Interface Interaction HCI) becomes the research hot spot of experts. Due to its natural and simple characteristic, gesture as a common adopted way of communicating aroused the attention of scholars. It is no doubt that introduces gesture into HCI will greatly enhance the user experience.The current gesture-based HCI research can be classified as four categories based on gesture modeling:data glove based approach, vision based approach, SEMG based approach and wearable sensor based approach. In the approaches above, vision based approach has higher accuracy and better interactive effect. But there is a problem difficult for it to deal with: occlusion, the user must be limited to the camera view with no obstructions. To solve the problem mentioned above, this dissertation presents a method of motion sensor based gesture recgonition, and attitude is added to the gesture model based on the previous works. Continues data stream collected by motion sensors are divied by threshold model, Ant Colony Algorithm are then introduced to do feature selection, finally HMM is used to recognize the gesture.The experimental verification is performed as well as the theoretical analyzing. An android phone (HTC HD2) is chosen as the data collecting device, and a prototype system based on HMM is developed to verify the effectiveness of the approach.20test gestures have been designed and used to do the recognition accuracy and comparison experiments. The results show that the approach can effectively recognize some gestures, especially complicated ones.