Human Activity Segmentation Challenge @ ECML/PKDD’23
Arik Ermshaus, Patrick Schäfer, Anthony Bagnall, Thomas Guyet, Georgiana Ifrim, Vincent Lemaire, Ulf Leser, Colin Leverger, Simon Malinowski
Code Available — Be the first to reproduce this paper.
ReproduceCode
Abstract
Time series segmentation (TSS) is a research problem that focuses on dividing long multivariate sensor data into smaller, homogeneous subsequences. This task is critical for various real-world data analysis applications, such as energy consumption monitoring, climate change assessment, and human activity recognition (HAR). Despite its importance, existing methods demonstrate limited efficacy on real-world multivariate time series data. To advance the field, we organized the Human Activity Segmentation Challenge at ECML/PKDD and AALTD 2023, featuring 57 participants. Collaborating with 15 bachelor computer science students, we gathered and annotated 10.7 h of real-world human motion sensor data. The challenge required participants to segment the resulting 250 multivariate time series into an unknown number of variable-sized activities. The top-8 approaches outperformed existing baselines, but show only limited improvements, capped at 1.9% points. The segmentation of real-world mobile sensing recordings remains challenging. We release the labelled challenge data for future research.