Sistem Deteksi Postur Duduk Ergonomis Menggunakan MOVENET dan LSTM

Azizi, Adam Haidar (2025) Sistem Deteksi Postur Duduk Ergonomis Menggunakan MOVENET dan LSTM. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211114-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (19MB) | Request a copy

Abstract

Penelitian ini mengembangkan sistem penilaian ergonomis otomatis yang mengintegrasikan teknologi estimasi pose MoveNet dengan jaringan saraf Long Short-Term Memory (LSTM) untuk memprediksi skor REBA pada postur duduk pekerja kantor. Metodologi meliputi ekstraksi keypoint berbasis MoveNet dari gambar dan video sudut samping untuk menghitung sudut sendi, diikuti dengan feature engineering komprehensif yang menggabungkan transformasi trigonometri, perhitungan logaritmik, turunan temporal, dan batasan biomekanik, dengan 30 fitur teratas dipilih berdasarkan analisis korelasi. Perbandingan arsitektur ganda antara model LSTM dan Bidirectional LSTM (BiLSTM) dilakukan menggunakan temporal data splitting, dan sequence preparation. Evaluasi performa menunjukkan bahwa LSTM dengan arsitektur yang lebih besar dari baseline (192-96-192 units) mencapai hasil terbaik secara keseluruhan dengan validasi R² = 0,9512 dan MAE = 0,0589, mengungguli baseline (160-80-160 units) sebesar 2,82% dalam R² dan 17,7% dalam MAE. Hyperparameter testing mengungkapkan bahwa LSTM memiliki sensitivitas tinggi terhadap optimasi dengan rentang performa luas (R² 0,9133-0,9512) dan mendominasi 5 dari 6 posisi teratas, sementara BiLSTM menunjukkan stabilitas lebih tinggi dengan rentang sempit (R² 0,9204-0,9452) namun margin peningkatan terbatas. Pengujian real-world menunjukkan tingkat keberhasilan 82,6% (19/23 skenario) dengan validasi ahli ergonomi menunjukkan sistem yang reliabel untuk penilaian ergonomis otomatis, meski memiliki limitasi pada deteksi gerakan 3D kompleks dan oklusi visual. Sistem berhasil mentranslasi postur berbasis computer vision menjadi skor risiko ergonomis yang relevan secara klinis, menyediakan kapabilitas monitoring otomatis dan berkelanjutan untuk intervensi kesehatan tempat kerja.
=====================================================================================================================================
This research develops an automated ergonomic assessment system integrating MoveNet pose estimation with Long Short-Term Memory (LSTM) neural networks to predict REBA scores for office worker sitting postures. The methodology involves MoveNet-based keypoint extraction from lateral-view images and videos to calculate joint angles, followed by comprehensive feature engineering incorporating trigonometric transformations, logarithmic scaling, temporal derivatives, and biomechanical constraints, with the top 30 features selected based on correlation analysis. A dual-architecture comparison between LSTM and Bidirectional LSTM (BiLSTM) models was conducted using temporal data splitting, and sequence preparation. Performance evaluation shows that LSTM with larger architecture than baseline (192-96-192 units) achieved the best overall results with validation R² = 0.9512, MAE = 0.0589, outperforming baseline LSTM (160-80-160 units) by 2.82% in R² and 17.7% in MAE. Hyperparameter testing revealed that LSTM exhibits high sensitivity to optimization with wide performance range (R² 0.9133-0.9512) and dominates 5 of 6 top positions, while BiLSTM demonstrates higher stability with narrow range (R² 0.9204-0.9452) but limited improvement margin. System validation across 23 scenarios achieved 82.6% success rate, with expert validation demonstrating reliable performance for automated ergonomic assessment, though limitations exist for postures involving twisting due to 2D pose estimation constraints. The system successfully translates computer vision-derived postural measurements into clinically relevant ergonomic risk scores, providing automated and continuous monitoring capabilities for workplace health interventions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Analisis Temporal, BiLSTM, Ergonomi, Estimasi Pose , Movenet, REBA, Ergonomic, Pose Estimation, Temporal Analysis
Subjects:	T Technology > T Technology (General) > T174.5 Technology--Risk assessment. T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.5 Information technology. IT--Auditing T Technology > T Technology (General) > T59.7 Human-machine systems.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Adam Haidar Azizi
Date Deposited:	27 Jul 2025 02:36
Last Modified:	27 Jul 2025 02:36
URI:	http://repository.its.ac.id/id/eprint/121668

Actions (login required)

View Item