E-Pen
Handwriting reconstruction using ML Time Series Classification
Overview
E-PEN is an experimental input system that captures handwriting through a one-dimensional pressure signal and reconstructs it into text using machine learning techniques. The project explores how much information density is actually required, with the result that surprisingly robust reconstructions are possible even with minimal data.
Challenge
The central challenge was the extremely limited data basis: a single sensor provides only one-dimensional time series data without any spatial information. To still achieve usable results, extensive data collection and the targeted use of contextual information were essential.
Concept & Implementation
Development of a pen equipped with an integrated pressure sensor and an interface for data acquisition and visualization. Based on extensive training data, a time series classification model was developed, enhanced with context-based post-processing. The system demonstrates that surprisingly high reconstruction performance can be achieved even under highly constrained conditions.
Showroom
