The Research On Gesture Recognition System Based On Inertial Sensors

With the rise of artificial intelligence,intelligent mobile devices include samrt phones,smart watches and smart hand rings are also seen everywhere in daily life.For the smart mobile devices,there are still some limitations in the touch screen interactive technology and voice interaction technology.Smart mobile devices have built-in inertial sensors,gesture recognition based on inertial sensors will be very suitable for the human-computer interaction of intelligent mobile devices.Gesture recognition,as a complement to the existing interactive method of intelligent mobile devices,can further improve the efficiency of human-computer interaction and improve the user experience.In this paper,we focus on gesture recognition system based on inertial sensors.In order to reduce the computational cost of data processing,we only use three axis accelerometer to recognize gestures.In the case of different gesture complexity,we propose different recognition models adapted to different application scenarios and develop a recognition system on smart phones,which achieves good recognition performance for simple control gestures and complex alphabet hand gestures.This paper mainly completed the following research work:1).An automatic gesture segmentation algorithm based on peak detection is proposed.By processing three axis accelerometer data,we get the reference signal.we detect peak value from the signal and use the threshold to automatically locate and segment the gesture motion.Finally we get the starting and ending point of the gesture motion.2).A generation model matching recognition scheme for simple control gesture recognition is proposed.The classification and regression tree(CART)algorithm is used to generate the regression tree model of all kinds of gestures,and the gestures are matched by the principle of the minimum square error.3).A discriminant model recognition scheme for complex gesture recognition is proposed.Gesture recognition method based on the reconstructed image from acceleration sensor signal,we use algorithm to convert sensor data to grayscale image.Then,the image is classified without artificial feature extraction.In addition,an undefined gesture filtering algorithm is added to the scheme.Finally,a simple and effective softmax regression model is used to classify gestures.