A Researchment Of Portable Electronic Device Gesture Recognition Technology Based On MEMS Inertial Sensors

With the rapid development of the mobile smart devices and the increasinglyimportant role it plays in people’s daily life, the natural and efficient human computerinterface become one of the hot field in the mobile smart device researchment. Due tothe low study effort and efficient interaction, touch screen and voice recognition replacethe keyboard which plays a dominant role in the smart devices. However, the twomethods also have disadvantages：the touch screen inevitably confines all the users’operation on the screen, the voice recognition is also helpless in the noisy environmentsuch as airport and bus station. Gestures have been employed in daily life since they arenatual to use and intuitive to be understood, so the gesture recognition based on MEMSinterial sensors could be occupied as a complementary interaction in the mobile smartdevice, making up the drawbacks of the touch screen and voice recogniton, enhancingthe interaction effeciency and user experience.This research project comes from UESTC-NOKIA international jointresearchment project, the main goal is to work out a simple and effective gesturerecognition algorithm which focus on the predefined gestures and based on MEMSinterial sensors. After investigating, analysing and comparing the prevalent methods ofthe gesture recognition based on MEMS interial sensors, finally this paper establishesits scheme with consideration to the premise of the researchment：1. As the Dynamic Time Warping and Hidden Markov Model recognition methods’poor adaptive capability of the difference among users, this paper proposes to extractthe features which reveal the gesture’s kinematic characters and patterns to recognizethe gestures, so the algorithm’s adaptive capability of the differences among uesrs isenhanced. Meanwhile, with the proposed hierarchy decision tree classifier, therecognition accuracy was improved significantly.2. The MEMS accelerometers are only occupied in the current approaches wichbased on feature extraction, the attitude information of the device couldn’t be acquired, so these algorithms only work well in the given device posture when the users operatethe gestures. As to this problem, this paper proposes to combine the MEMSaccelerometer and gyroscope to get the device attitude information, then transform thegesture data from the device coordinate to the user’s coordinate before gestruerecogniton, this eliminates the limination to the device posture of gestures and enablesthe gesture recignition in any postures.3. As the device attitude updating algorithm’s high requirement of the dataprecision, this paper calibrates the MEMS accelerometer and gyroscope respectivelyaccording to the system error models, sets up the time series model of the sensor’srandom drift and then eliminates with the Kalman filter.