Detecting Writing Micro-Events using Motion Sensors in Smartwatches

Handwriting is integral to the education system, where students write notes, assignments, and exams using handwritten words. We explore the potential of motion sensors in a smartwatch to understand the writing micro-events — hand shift, line change, and actual writing. The ability to detect these writing micro-events opens up different possible studies, such as studying the students' cognitive load, comparing students of a different or similar caliber, their familiarity with given content, and cheating during examinations. Unlike existing work on handwriting using commercial smartwatches, which focuses on writing content or writer detection, we focus on detecting the writing micro-events. In our offline evaluation with data from ten participants, we observe F1 scores up to 0.84 for writing conditions of copying, writing from memory, and writing original text. A user-specified sample increases the best F1 score to 0.91.