A handwritten character recognition system based on acceleration

In this paper, an handwritten character recognition system based on acceleration is presented. The character recognition system using a 3-dimensional (3D) accelerometer, includes three procedures: original signal detection, signal processing (preprocessing and quantization) and recognition/classification. In quantization procedure, Trajectory Orientation (TO) and Curve Feature (CF) are adopted and compared. In recognition procedure, Fully-connected Hidden Markov Model (HMM) and Left-Right HMM are both implemented and compared. The system, in the recognition of 10 Arabic numerals, achieves the Correct Rate(CR) of 99.05% and the Total Correct Rate (TCR) of 94.76%.