Air Write Letter Recognition Using Random Forest Classification on Arduino Dataset

Abstract

In this paper, we propose to use a random forest technique based on a bundle of decision tree classifiers as weak classifiers to classify the Turkish letters written on the air. Air write recognition enable applications to read the text that users wrote on the air as an invisible surface. Such applications are very useful for disabled people. Air writing here uses Arduino MPU 6050 sensors, namely gyroscope and accelerometer. As a main contribution, random forest classification is applied to air write recognition problem using the Arduino dataset for Turkish letters. Features are extracted from sensor signals using 2-dimensional Fourier transform. The results show that Random Forest outperforms the other three methods, namely k-nearest neighbor, decision tree and Mahalonobis classifier.

Key Words: Air write, letter recognition, Arduino, Fourier transform, Random forest, Decision tree