TOWARDS AI-BASED AT-HOME PHSIOTHERAPY FOR THE ELDERLY: COMPARING MOVENET WITH VITPOSE IN HUMAN POSE ESTIMATION

Authors

Dominik Batić, Ostfalia University of Applied SciencesFollow
Srinivas Kachavarapu, Ostfalia University of Applied SciencesFollow
Fayçal Nait Irahal, Ostfalia University of Applied SciencesFollow
Dagmar Meyer, Ostfalia University of Applied SciencesFollow
Carsten Meyer, Ostfalia University of Applied SciencesFollow

Category

Rehabilitation

Document Type

Paper

Abstract

The current tendency to shorten hospital stays requires patients to do rehabilitation exercises independently at home after knee and hip replacement surgeries. To help address the lack of supervision, we are developing a digital system that uses Human Pose Estimation (HPE) to analyse patient motion and provide feedback. In this work, we focus on two system aspects: (1) A tablet-based mobile application with an elderly-friendly interface and (2) a quantitative comparison of a widely used real-time HPE algorithm (MoveNet) with one of the top performers on the COCO benchmark (ViTPose). We provide experimental results for two-dimensional (2D) HPE on public FreeMan single-person data. Compared to MoveNet, ViTPose achieves almost 19% higher Percentage of Correct Keypoints (PCK) at a 5% torso threshold and nearly 8% higher PCK at a 10% threshold.

Recommended Citation

Batić, Dominik; Kachavarapu, Srinivas; Nait Irahal, Fayçal; Meyer, Dagmar; and Meyer, Carsten (2025) "TOWARDS AI-BASED AT-HOME PHSIOTHERAPY FOR THE ELDERLY: COMPARING MOVENET WITH VITPOSE IN HUMAN POSE ESTIMATION," ISBS Proceedings Archive: Vol. 43: Iss. 1, Article 9.
Available at: https://commons.nmu.edu/isbs/vol43/iss1/9