Motion Is All You Need

The integration of digital tablets and smart pens has revolutionized handwriting tasks in various domains, yet current solutions like the Apple Pencil necessitate specialized touchscreens, escalating costs and limiting accessibility. This study introduces a novel motion-based handwriting recognition system that eliminates the dependency on touchscreens by utilizing motion data alone, enabling handwriting capture on any surface. Our primary contributions include the design of an all-in-one, 3D-printed smart pen prototype equipped with an inertial measurement unit (IMU) for precise motion sensing, and the creation of a comprehensive handwriting motion dataset comprising 6,500 samples of lowercase English characters. We developed end-to-end machine learning algorithms, specifically a 1D Convolutional Neural Network (1D CNN) and a Long Short-Term Memory (LSTM) network, to accurately recognize handwritten characters from the captured motion data. Experimental results demonstrate that our LSTM model achieves a test accuracy of 90.46%, significantly outperforming previous motion-based recognition systems. This work validates the feasibility of motion-only handwriting recognition and presents a cost-effective alternative to touchscreen-dependent smart pens, potentially broadening the accessibility of digital handwriting technologies.