Upper Limb Movements Recognition System Based On Sensor

As the development of computer technology, pattern recognition technology and sensor technology, Accelerometer-based human motion pattern recognition as an emerging field of pattern recognition research direction, the actual acceleration sensor is collected through the use of human body movement acceleration data, and then the signal preprocessing, feature extraction and selection and, ultimately, for classification of the action. Since the development of MEMS technology, there are many domestic and foreign-based acceleration sensor consumer electronics products have achieved great success, including Accelerometer-based human-computer interactive games, popular products, bringing users a new experience to subvert the traditional game concept.Accelerometer-based human motion recognition and interactive game has a huge advantage, but due to the complexity of the surrounding environment and the diversity of human motion human motion recognition makes it relatively difficult and there are some flaws, the current research in the field still at a relatively basic stage. In Japan, Nintendo’s "WiiSports" game, players make only slight jitter in the game content is displayed on a huge range of action, due to technical reasons can not accurately identify the user’s actions, for example in table tennis even for several common forehand, backhand, smash such action can not be accurately identified. This defect causes the reason is because the upper limb acceleration access to information and data processing methods properly, a direct result of these defects content in the game can not be accurately restore the user’s actions, allowing users to experience a sense of the game is not true.To solve this problem, we made the following tasks:1) in the human body model is proposed based on the analysis for human-computer interactive games human upper limb motion signal acquisition program, to build the necessary hardware acquisition platform2) the design of a suitable interactive games simple common table tennis forehand, backhand, smash hit three kinds of action classification and recognition algorithm,3-way and3-way acceleration signal signal signal pre gyro after treatment, using wavelet transform method for feature extraction, and then uses a combination of PAC and ReliefF feature selection feature selection algorithm is complete, the final establishment of SVM model classification.3) designed by the experiments on20subjects operation signal samples were collected, each subject experiment was repeated four times, each time making repeated experiments mimic forehand table tennis players, backhand smash hit ball action, each action do it10times, and has been more satisfactory results.