Machine Learning for Sensor-Based Handwritten Character Recognition: A Brief Survey

Abstract

Handwriting recognition technologies have evolved significantly, with recent advancements incorporating wearable devices such as smartpens, smartwatches, and smartphones to capture motion and gesture data. This paper reviews various efficient handwriting recognition models that utilize inertial sensors, visual sensors, EMG sensors, and acoustic and Wi-Fi signals. Inertial sensor-based approaches often use accelerometers and gyroscopes to collect three-dimensional data during handwriting, with models achieving high accuracy in recognizing digits and alphabets. Visual sensor-based techniques, leveraging cameras and image processing, have been fundamental in developing Optical Character Recognition (OCR), further enhanced by deep learning algorithms. EMG sensors have been explored for their potential to capture muscle activity for gesture recognition. At the same time, acoustic and Wi-Fi signal-based methods offer innovative ways to recognize handwriting through sound and signal variation. Despite the progress, challenges such as sensor drift, data generalization, and computational requirements persist. Integrating multi-modal sensors and advanced machine learning techniques holds promise for overcoming these limitations and improving the accuracy and usability of handwriting recognition systems across diverse applications. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.