Research Of Inertial Signal-based Handwritten Numeral Recognition

Compared with the traditional way of the mouse, keyboard, touch screen input, handwritten pens,as a new man-machine input mode,are more in line with the habits of people and make man-machine input more liberalized. In this paper, we have studied the handwritten digit recognition based on inertial sensor, through the analysis of inertial signals which collected in the process of user’s handwriting of numbers to identify the current numbers written by the user. This paper’s goal aims to achieve a robust, low computational complexity of handwritten numeral recognition method, which is easy to transplant to hardware stylus processing hardware that are low power consumption and low cost.There are two kinds of different ideas about handwritten digit recognition based on inertial navigation signal. One is based on the digital identification of complete sequence of source signals. Another is based on the digital recognition of the characteristics of sequences. In order to reduce computational complexity, this paper proposes a handwritten numerals recognition method based on the characteristics of sequence, the ideas and contributions of this paper are as follows:(1) We presents a desktop handwritten behavior identification method based on inertial navigation signal, using handwritten pen and the contact surface friction in the process of high frequency vibration as the instruction, and through the energy detection and the zero test to extract the signal section in the out.(2) We proposed a kind of handwritten acceleration signal adaptive multi-dimensional characteristics of peak valley feature extraction method, which significantly reduce data complexity of identify object.(3) For a small amount of digital identification of fuzzy problems under the above characteristics, we further put forward two axes inertial navigation signal peak valleyphase difference as the characteristics of secondary identification, so as to ensure the accuracy of identification for the figures.(4) We find it is difficult to distinguish number 0 and 6 for the acceleration signals. However we discover one identification method based on the endpoint attitude angle characteristics which can effectively distinguish between digital 0 and 6.In the experimental stage,we compare the method in this paper with several other methods from the identification accuracy and the time complexity aspect. The experimental results show that the proposed handwritten numeral recognition method on the recognition accuracy and time complexity is superior to other methods.