Development of Inertial Sensor Based Handwritten Digit Trajectory Reconstruction Algorithm

• Main Authors: Jiun-Nan Liu, 劉俊男
• Other Authors: Jeen-SHin Wang
• Format: Others
• Language: en_US
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/82917206044458843626

碩士 === 國立成功大學 === 電機工程學系碩博士班 === 97 === This thesis presents an inertial-measurement-based pen (IMUPEN) and a trajectory reconstruction algorithm. The trajectory reconstruction algorithm is capable of providing reconstructed trajectories on demand with the IMUPEN, composed of a triaxial accelerometer, two gyroscopes, a microcontroller, and an RF transmitter. The IMUPEN alone with the trajectory reconstruction algorithm can be used for off-desktop human computer interaction (HCI). The trajectory reconstruction algorithm comprises two constituent procedures: an attitude estimation procedure and a position estimation procedure. The signals collected from the inertial sensors are preprocessed before the trajectory reconstruction algorithm uses them. The signal preprocessing procedure includes a sensor calibration step and a low-pass filtering step. The attitude estimation employs both the preprocessed accelerations and angular velocities to produce attitude estimates. A Kalman filter and a switching mechanism have been developed and integrated into the attitude estimation procedure to make estimation more accurate. The position estimation procedure produces position estimates using the preprocessed accelerations and the attitude estimates. A multi-axes dynamic switch (MAD) has been developed to reduce position drift from double integration of accelerations. The accuracy of reconstructed trajectories is satisfactory for short-term motion trajectory reconstruction and handwritten digit recognition.