Air Writing via Receiver Array-Based Ultrasonic Source Localization

• Published in: IEEE transactions on instrumentation and measurement 2020-10, Vol.69 (10), p.8088-8101
• Main Authors: Chen, Hui, Ballal, Tarig, Muqaibel, Ali Hussein, Zhang, Xiangliang, Al-Naffouri, Tareq Y.
• Format: Article
• Language: English
• Subjects: Accuracy; Acoustic localization; Acoustic waves; Acoustics; air writing; Algorithms; Classification; Classifiers; Direction of arrival; direction of arrival (DOA); Direction-of-arrival estimation; Estimation; Forecasting; Hidden Markov models; human–machine interaction; Machine learning; phase difference; receiver arrays; Receivers; Recognition; Sensor arrays; sequence classification; Text recognition; Tracking; Training; Writing
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2020.2991573

Air-writing systems have recently been proposed as tools for human-machine interaction where instructions can be represented using letters or digits written in the air. Different technologies have been used to realize air-writing systems. In this article, we propose an air-writing system using acoustic waves. The proposed system consists of two components: a motion-tracking component and a text recognition component. For motion tracking, we utilize direction-of-arrival (DOA) information. An ultrasonic receiver array tracks the motion of a wearable ultrasonic transmitter by observing the change in the DOA of the signals. We propose a novel 2-D DOA estimation algorithm that can track the change in the direction of the transmitter using measured phase differences between the receiver array elements. The proposed phase-difference projection (PDP) algorithm can provide accurate tracking with a three-sensor receiver array. The motion-tracking information is passed next for text recognition. To this end, and in order to strike the desired balance between flexibility, processing speed, and accuracy, a training-free order-restricted matching (ORM) classifier is designed. The proposed air-writing system, which combines the proposed DOA estimation and text recognition algorithms, achieves a letter classification accuracy of 96.31%. The utility, processing time, and classification accuracy are compared with four training-free classifiers and two machine learning classifiers to demonstrate the efficiency of the proposed system.