Gesture Pattern Recognition Based On MEMS Interial Sensors

Nowadays, as electronic mobile devices are developing rapidly, people’s life associates with those devices more and more closely. Man-machine interaction sets up a communicating bridge between users and devices, and the key methods of this communication are through key boards, touch screen and voice recognition. Nevertheless, constrained by the size of mobile devices, users` utility from using keyboard operation is limited. Meanwhile, the accuracy rate of voice recognition is not high enough neither, as its accuracy is greatly affected by gender, speaking speed, pronunciation accuracy, ambient noise, microphone quality and so on. For these reasons, this article proposes a new method which is a MEMS interial sensors based gesture pattern recognition. The essence of this method is that, it collects the original data of gestures through interial sensors and transfers all these data into computers or phones to be processed; after that, it extracts features of different gestures and finally using classifier to classify and distinguish them. So far, there are five key research methods, which are Hidden Markov Model, Dynamic Time Warping, Fuzzy Neural Network, Path Reconstruction and The Original Gesture Data Statistics Analysis, and this article focuses on the last method.Given that MEMS interial sensors based gesture pattern recognition model is still in its primary stage, study of this topic still faces numbers of difficulties, such as, extraction of feature quantity, design of classifier, and many more. For this reason, the purpose of this article is to resolve some of these difficulties, and the main focuses of this article will be:(1) Run error analysis on MEMS accelerometer and gyroscope; presents a calibration for error model of MEMS interial sensors; and using the filter to filter random error of sensors in order to assure accuracy of algorithm.(2) Since individual differences of extracted feature quantity is minimized during the process of extracting it, this enables recognition algorithm to be unaffected by individuals, hence, this method has been widely adopted. More than that, the use of coordinate transformation, attitude updating and gyroscope enables gyroscope to be less strict on device gesture. Consequently, users are able to accurately recognize gestures in any circumstances, and this vastly improved the application range of this algorithm.(3) Classifier is designed based on the extracted feature quantity, and this classifier can accurately identify the four key gestures. As well as that, it is also able to identify directions and frequencies of different sub-gestures of those four gestures. Combining these with accelerometer and gyroscope, all types of sub-gestures can be accurately classified.