Hand-written character recognition using MEMS motion sensing technology

In this paper, a micro inertial measurement unit (muIMU) based on micro electro mechanical systems (MEMS) sensors is applied to sense the motion information produced by characters written by human subjects. The muIMU is built to record the three-dimensional accelerations and angular velocities of the motions during hand-writing. (Here we write the characters in a plane, so only two accelerations and one angular velocity are taken from muIMU in processing the data discussed in this paper). Then, we compared the effectiveness of data processing methods such as FFT (fast Fourier transform) and DCT (discrete cosine transform) by showing their corresponding experimental results. Subsequently, we gave an analysis of these two methods, and chose DCT as the preferred data processing method. For character recognition (26 English alphabets and 10 numerical digits), unsupervised network self-organizing map (SOM) is applied to classify the characters and comparatively good results are obtained. Our goal is to show the feasibility of character recognition based on selected sensor motion information, and provide a potential technology for human-gesture recognition based on MEMS motion sensors.