Estimasi Pose 2D Berbasis Skeleton Untuk Identifikasi Titik Sudut Penting dan Koreksi Postur Pada Latihan Beban Dasar

2D Pose Estimation Based on Skeleton for Identifying Key Joint Angles and Posture Correction in Basic Weight Training 

Perubahan gaya hidup modern telah menyebabkan peningkatan penyakit tidak menular seperti hipertensi, obesitas, dan diabetes melitus. Latihan kekuatan terbukti efektif dalam mencegah dan mengendalikan penyakit-penyakit tersebut, namun pelaksanaannya sering terkendala akses terhadap pelatih atau fasilitas kebugaran. Penelitian ini mengembangkan sistem pelatihan otomatis berbasis pose estimation dua dimensi menggunakan MediaPipe, yang mampu mendeteksi posisi sendi dan menghitung sudut gerakan secara real-time. Sistem ini mengenali tiga jenis latihan beban dasar—bicep curl, squat, dan lateral raise—dengan memberikan umpan balik terhadap kesesuaian postur berdasarkan standar gerakan optimal. Hasil pengujian menunjukkan tingkat akurasi tinggi: bicep curl ≥96% di semua sudut kamera, lateral raise mencapai 97,33%, dan squat tetap akurat meskipun sedikit menurun pada sudut 90°. Evaluasi menggunakan Mean Absolute Error (MAE) menunjukkan bahwa sudut kamera berpengaruh signifikan terhadap deviasi sudut gerakan. Mayoritas partisipan berada dalam kategori deviasi minor (<15°), meskipun beberapa subjek perempuan menunjukkan nilai MAE >15° pada squat dan bicep curl. Sistem ini terbukti efektif memberikan umpan balik korektif dan berpotensi digunakan dalam pelatihan mandiri yang aman dan adaptif.

Modern lifestyle changes have led to a rise in non-communicable diseases such as hypertension, obesity, and diabetes mellitus. Strength training has proven effective in preventing and managing these conditions; however, its implementation is often hindered by limited access to trainers or fitness facilities. This study developed an automated training system based on two-dimensional pose estimation using MediaPipe, capable of detecting joint positions and calculating movement angles in real time. The system is designed to recognize three basic strength exercises—bicep curl, squat, and lateral raise—while providing feedback on posture accuracy compared to optimal movement standards. Test results show high detection accuracy: bicep curl ≥96% across all camera angles, lateral raise reached 97.33%, and squat maintained strong performance despite a slight decrease at the 90° side view. Evaluation using Mean Absolute Error (MAE) revealed that camera angle significantly affects motion deviation. Most participants fell within the minor deviation category (<15°), although some female subjects recorded MAE values >15° in squat and bicep curl exercises. The system effectively provides corrective feedback and has strong potential for safe, adaptive, and independent strength training.