Journal DOI 10.68956/journal.human-performance-futures

Creative Commons source reading · Safeguarding and athlete safety

Deep learning-based physical exercise assessment of older adults using single-camera videos

Vayalet Stefanova; Evelien Maeyens; Jeroen Brughmans; Dimitri Vargemidis; Julien Lebleu; Leen Stulens; Sanne Broeder; Karen Gilis; Mieke Deschodt; Benedicte Vanwanseele; David Beckwée; Moran Gilat; Benjamin Filtjens; Bart Vanrumste

Originally published in PLOS Digit Health · 2026-08-06

From the article

Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.

Abstract

Regular physical activity preserves functional independence in older adults, yet care-home residents often miss out because personalized supervision is scarce. Autonomous, technology-supported exercise platforms could deliver such guidance without additional staff time—but only if sessions are automatically monitored for safety and quality. We therefore designed a deep learning (DL) system that (a) recognizes individual exercise types and (b) estimates joint angle trajectories from a standard video recording. These outputs are used to compute objective exercise performance metrics (EPMs) such as duration, repetition count, motion variability, and range of motion. Seven care-home residents (aged between 65–94 years) performed six common rehabilitation exercises in front of a single camera while wearing 17 inertial sensors (Xsens MVN…

Results

This section begins with a description of the data characteristics, followed by an overview of the performance of the implemented DL model on (1) exercise recognition and (2) joint angle prediction. Finally, the accuracy of EPMs estimated from model predictions is compared to those derived from ground truth data.

Discussion

This study presents a novel method for the unsupervised monitoring of older adults’ exercise performance using a single-camera system combined with deep learning. We demonstrated that a DL model can be trained to successfully recognize exercise actions and estimate joint angles in older adults. Our implementation relied on existing tools: AlphaPose for 2D pose estimation and VideoPose for processing temporal pose data for both classification and regression tasks. While the system showed promising performance, future work could explore whether alternative 2D pose estimators, such as OpenPose [65] and PoseNet [66], or different model architectures might improve accuracy. However, it is important that any future improvements take into account the computational power required to run these algorithms to ensure that their deployment in…

Editorial reading guide

  • Was the study designed for the population to which its conclusion is applied?
  • What safeguard, access need or equity consequence should shape implementation?