Handwritten character recognition in the air by using leap motion controller

Abstract

In order to develop a system which can precisely and quickly recognize handwritten characters in the air by using a Leap Motion Controller, we propose the following method: (1) A user has to register handwritten characters as template patterns before use. Each pattern is represented by a sequence of motion vectors calculated by using adjacent sampling data. (2) In the recognition phase, an input pattern is represented in the same method as above. The input pattern is compared with each of the registered template patterns by using DP matching and we can obtain a distance (degree of similarity) between them. Our system outputs the character class corresponding to the pattern with a minimum distance as a recognition result. In our experiments for recognition of 46 Japanese hiragana characters and 26 alphabets, a high average recognition rate of 86.7 % and a short average processing time of 196 ms were obtained.