Sistem Deteksi dan Evaluasi Latihan Fisik Berbasis Mediapipe Blazepose dan Finite State Machine pada Raspberry Pi

Haqq, Zahid Miftakhul (2026) Sistem Deteksi dan Evaluasi Latihan Fisik Berbasis Mediapipe Blazepose dan Finite State Machine pada Raspberry Pi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022221018-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Diperkirakan sekitar 50% dari semua cedera olahraga dapat dicegah dengan teknik yang tepat dan intervensi pelatih. Namun aksesibilitas dan biaya menjadi keterbatasan. Penelitian ini bertujuan untuk mengembangkan sistem berbasis MediaPipe Human Pose Estimation (BlazePose) yang diimplementasikan pada Raspberry Pi 4 untuk menghitung repetisi dan mengevaluasi kualitas gerakan latihan push-up dan squat secara real-time. Sistem menggunakan dua webcam dari sudut samping dan depan untuk menangkap landmark tubuh, yang diproses melalui pipeline threading (ThreadedCapture dan PoseWorker) dengan sinkronisasi timestamp. Penghitungan repetisi dilakukan dengan merepresentasikan gerakan sebagai motion signal yang dihaluskan dengan Exponential Moving Average (EMA) dan dibandingkan terhadap baseline adaptif time-corrected melalui Mealy Finite State Machine (FSM). Evaluasi kualitas gerakan dilakukan dengan menormalisasi error tiap metrik biomekanik tubuh terhadap rentang ideal, lalu diagregasi dengan Weighted Root Mean Square Error (WRMSE). Hasil pengujian menunjukkan bahwa model complexity memengaruhi inferensi jauh lebih besar dibandingkan resolusi citra, dan frame rate (FPS) memiliki korelasi kuat (Pearson r = 0.93) terhadap tingkat deteksi repetisi. Konfigurasi optimal pada Raspberry Pi diperoleh pada model complexity Lite dan resolusi 480x270, dengan akurasi deteksi repetisi 81.67%, akurasi skoring kualitas gerakan 72% (MCC = 0.453), serta rata-rata latensi sinkronus 89 ms pada implementasi real-time. Penelitian ini membuktikan bahwa edge computing berbasis Raspberry Pi dan MediaPipe dapat digunakan untuk membangun sistem pelatih kebugaran otomatis.
=====================================================================================================================================
It is estimated that approximately 50% of all sports injuries can be prevented with proper technique and coaching intervention. However, accessibility and cost become limitations. This research aims to develop a system based on MediaPipe Human Pose Estimation (BlazePose) implemented on a Raspberry Pi 4 to count repetitions and evaluate the movement quality of push-ups and squats in real-time. The system uses two webcams from side and front angles to capture body landmarks, which are then processed through a threading pipeline (ThreadedCapture and PoseWorker) with timestamp synchronization. Repetition calculations are performed by representing movements as a smoothed motion signal with an Exponential Moving Average (EMA) and compared against an adaptive time-corrected baseline using a Mealy Finite State Machine (FSM). Evaluation of movement quality is carried out by normalizing the error of each body biomechanical metric to the ideal range, then aggregated with Weighted Root Mean Square Error (WRMSE). The results show that model complexity influences inference much more than image resolution, and frame rate (FPS) has a strong correlation (Pearson r = 0.93) with the repetition detection rate. The optimal configuration on the Raspberry Pi was obtained with the Lite complexity model and a resolution of 480x270, with a repetition detection accuracy of 81.67%, quality scoring accuracy of 72% (MCC = 0.453), and an average synchronous latency of 89 ms on the real-time implementation. This study proves that Raspberry Pi-based edge computing can be used to build an automated fitness trainer system.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Raspberry Pi, MediaPipe, Edge Computing, Exponential Moving Average, Finite State Machine.
Subjects:	T Technology > T Technology (General) > T385 Visualization--Technique T Technology > T Technology (General) > T58.62 Decision support systems T Technology > T Technology (General) > T59.7 Human-machine systems.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Zahid Miftakhul Haqq
Date Deposited:	21 Jul 2026 02:32
Last Modified:	21 Jul 2026 02:32
URI:	http://repository.its.ac.id/id/eprint/135807

Actions (login required)

View Item