Shizuoka University
Okabe Lab
日本語

Personalized Posture Evaluation Method
Using Skeletal Information and an LLM

Rikuto Katsura

Shizuoka University

Makoto Okabe

Shizuoka University

研究のTeaser画像

System Overview Diagram. Skeletal information is extracted from webcam footage using MediaPipe and sent to ChatGPT for posture evaluation.

Abstract

In this study, we propose a system that allows the user to register their ideal posture as a reference and, by comparing it with their current posture every 30 minutes, detects and points out changes or deterioration in posture.When the system is launched, the webcam is activated. The user first sits in their ideal posture and presses the Enter key. Then, using MediaPipe, skeletal information is extracted solely from the webcam image and saved in JSON file format. After 30 minutes, the skeletal information of the current posture is recorded again. The saved skeletal data is then sent to ChatGPT, which outputs a posture evaluation and suggestions for improvement. This comparison between the current posture and the ideal posture is repeated at 30-minute intervals thereafter.

Paper

Bachelor Thesis

IPSJ SIG-CGVI, 199th conference

Video

BibTeX citation

    @Misc{katsura.rikuto.20_bachelor,
        author = {Rikuto Katsura},
        title = {Personalized Posture Evaluation Method Using Skeletal Information and an LLM},
        howpublished = {Bachelor Thesis},
        month = {8}
        year = {2025},
    }