Gesture Recognition And Application To Human-computer Interaction System Based On Surface Myoelectric Signals

The surface myoelectric signal contains rich information about human motion and the gesture recognition based on myoelectric signal with applications in mechanical control systems has been attracting an ever-increasing research interest in recent years.Nevertheless,there are still problems such as large delays,low recognition rate due to the complexity of myoelectric signals,and thereby leading to the motivations for this thesis to develop a practical human-computer interaction system by using gesture motions controlling the quadrotor.The main contents are as follows:First,the gesture recognition techniques are studied based on myoelectric signals in three aspects,namely,the data segmentation methods,feature analysis algorithms and classifier approaches.The transient energy threshold is introduced to determine the active signal segment of the raw surface myoelectric signals.As a result,the convenience of label making is improved.In addition,the appropriate characteristics of myoelectric signal are selected and static classification models are established by linear discriminant analysis.Experiments are conducted with respect to the classification precision and rapidity.Second,an adaptive gesture recognition system is proposed based on the incremental linear discriminant analysis,which aims at solving the problem that the static gesture classifier for a specific user cannot adapt to other users.By using the idea of incremental learning,the general classification model is dynamically updated to an adaptive one that can be used for other users.Experiments are given to show the advantages of the proposed algorithms over the general one.Third,a real-time gesture recognition and interactive system is launched based on myoelectric signals.With the purpose of reducing the high error rate of the interactive experiment,a new mechanism is designed by combining the acceleration feature information.Meanwhile,a classifier is built with new feature vectors by fusing two channels myoelectric signals and acceleration signals.As such,the calculation time is reduced and the recognition accuracy is improved.In addition,a friendly human-computer interaction interface is implemented and a communication platform is constructed with the quadrotor aircraft system.Then,the interaction can be realized by sending the recognition results to the quadrotor and generating the control commands corresponding to certain gestures.