Air Writing Recognition using Machine Learning Algorithms

Air Writing Recognition is one of the most innovative forms of human and computer interaction. Recognizing Gesture based writing in Air helps in the analysis of hand movements without touchpad or screen to trace and to convert them into written or digital text images. Air writing recognition system is developed using sensors that helps to recognize the characters with the help of accelerometer and gyroscope data. Air written characters face challenges in the writing styles of participants, the articulation speed and thereby exhibits difficulty in effective character writing. Existing research works to recognize air written characters have been carried out using CNN and LSTM with captured images. The proposed methodology suggests an improved Air Writing Recognition system with a smart band worn in the wrist. The data collected using the smart band is wirelessly transferred using the Bluetooth module. Three Machine Learning algorithms like RF, KNN, GBM were trained using the acquired data. The performance of the machine learning model was compared using the metrics like Accuracy, Precision, Sensitivity and Specificity and Recall. The accuracy of the KNN model is found to be better than the other two algorithms for the digits. Simulation results show that the accuracy of the KNN model is .49%, 27.57% higher than RF and GBM model respectively.