Sistem Monitoring Postur Duduk Pada Pekerja Sedenter Berbasis Visi Komputer Dan Raspberry Pi

Haidi, Farih Labib (2026) Sistem Monitoring Postur Duduk Pada Pekerja Sedenter Berbasis Visi Komputer Dan Raspberry Pi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022221121-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (81MB) | Request a copy

Abstract

Postur duduk yang tidak netral dan dipertahankan dalam waktu lama dapat meningkatkan risiko gangguan muskuloskeletal pada pekerja sedenter. Penelitian ini bertujuan merancang dan mengevaluasi secara operasional prototipe monitoring postur duduk berbasis visi komputer yang berjalan secara lokal pada Raspberry Pi. Sistem menerima masukan kamera atau video, mendeteksi pengguna pada region of interest (ROI) kursi, dan memperkirakan landmark tubuh menggunakan MediaPipe Pose. Landmark tersebut digunakan untuk menghitung proksi craniovertebral angle (CVA), trunk lean, rasio kepala maju, dan hunch index. Setelah penyaringan temporal, aturan heuristik berbasis ambang menentukan status setiap frame sebagai GOOD atau POOR. Sistem menghasilkan anotasi visual, video MP4, log CSV per frame, dan pemicu peringatan ketika status POOR berlangsung berkelanjutan. Pengujian dilakukan menggunakan sepuluh video pengguna dengan durasi sekitar 300 detik per video. Sistem menghasilkan 14.262 baris pose, yang terdiri atas 8.004 baris GOOD dan 6.258 baris POOR. Kolom voice_alert mencatat 71 pemicu peringatan, sedangkan kategori masalah yang paling sering terdeteksi berkaitan dengan leher atau kepala. Hasil tersebut menunjukkan bahwa alur monitoring dan pencatatan dapat dijalankan serta ditelusuri pada data uji. Namun, distribusi status merupakan keluaran sistem dan belum menunjukkan akurasi karena video belum memiliki label ground truth manual pada frame dan subjek yang sama. Validasi lanjutan memerlukan data berlabel dan pengujian keluaran audio secara langsung.
==================================================================================================================================
A non-neutral sitting posture maintained for prolonged periods may increase the risk of musculoskeletal disorders among sedentary workers. This study aims to design and operationally evaluate a computer vision-based sitting-posture monitoring prototype that runs locally on a Raspberry Pi. The system accepts camera or video input, detects a user within the chair region of interest (ROI), and estimates body landmarks using MediaPipe Pose. These landmarks are used to calculate a proxy craniovertebral angle (CVA), trunk lean, forward head ratio, and hunch index. Following temporal filtering, threshold-based heuristic rules classify each frame as GOOD or POOR. The system produces visual annotations, MP4 video, per-frame CSV logs, and a warning trigger when a POOR status persists. Testing used ten user videos, each approximately 300 seconds long. The system produced 14,262 pose rows, comprising 8,004 GOOD rows and 6,258 POOR rows. The voice_alert column recorded 71 warning triggers, while the most frequently detected problem category concerned the neck or head. These results show that the monitoring and logging workflow can operate on the test data and that its decisions can be traced through recorded features. However, the status distribution represents system output rather than accuracy because the videos do not have manual ground-truth labels for the same subjects and frames. Further validation requires labeled data and direct testing of the audio output.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Postur Duduk, Pekerja Sedenter, Visi Komputer, MediaPipe Pose, Algoritma Heuristik, Raspberry Pi. Sitting Posture Monitoring, Sedentary Workers, Computer Vision, Human Pose Estimation, MediaPipe Pose, Heuristic-Based Classification, Raspberry Pi.
Subjects:	Q Science > QP Physiology > QP34.5 Human engineering
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Farih Labib Haidi
Date Deposited:	24 Jul 2026 02:48
Last Modified:	24 Jul 2026 02:48
URI:	http://repository.its.ac.id/id/eprint/136677

Actions (login required)

View Item