Air-Writing Character Recognition with Ultrasonic Transceivers.

Area of Science:

• Human-Computer Interaction
• Signal Processing
• Machine Learning

Background:

• User interfaces are evolving towards natural communication methods.
• Air-writing is an emerging gestural interaction technique.
• Existing systems may require complex sensor setups.

Purpose of the Study:

• To propose a new air-writing system utilizing a single array of ultrasonic transceivers.
• To evaluate the performance of various deep learning algorithms for air-written character recognition.
• To demonstrate a high-performance, practical air-writing solution.

Main Methods:

• A single array of ultrasonic transceivers was used to capture hand marker track data based on pairwise distances.
• Deep learning models including Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), Convolutional Autoencoder (ConvAutoencoder), and Convolutional LSTM (ConvLSTM) were trained and evaluated.
• The system was validated using real-world data acquisition and classification scenarios.

Main Results:

• ConvLSTM achieved the highest character recognition accuracy at 99.51%.
• The system demonstrated low latency, with ConvLSTM achieving 71.01 milliseconds.
• The proposed system showed high performance in both data acquisition and classification in real scenarios.

Conclusions:

• A single ultrasonic transceiver array is sufficient for an effective air-writing system.
• Deep learning, particularly ConvLSTM, is highly effective for recognizing air-written characters.
• The developed system offers a practical and high-performance solution for natural gestural interaction.