Title: Identifying Pen-In-Air States via Video: A Proof-of-Concept for Supplementary Handwriting Analysis

Abstract:

The dynamic components of handwriting play a vital role in evaluating developmental conditions like dysgraphia, with digitizing tablets currently serving as the standard capture method. However, relying on tablet-based sensors limits the analysis of Pen-Up behavior to a narrow proximity zone just above the writing surface, which may cause researchers to overlook significant high-lift movements occurring in the air. To address this limitation, this study serves as a proof of concept, exploring whether top-view video can offer complementary data for determining pen-contact states without depending on tablet proximity sensors. We introduce an interpretable hybrid framework that integrates kinematic feature extraction and machine learning classification with pen-tip tracking, utilizing a detector based on YOLO. For validation, a pilot dataset comprising various handwriting videos was manually annotated at the frame level, and performance was assessed using a Leave-One-Video-Out (LOVO) protocol. The approach demonstrated robust event-level detection of Pen-Up segments, achieving an F_2 score of up to 0.805. This metric aligns with the high recall requirements typical of screening contexts. These findings confirm the viability of video-based Pen-Up detection as an affordable, non-intrusive addition to digitizing tablets, laying the groundwork for subsequent large-scale research initiatives.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC