Analisis Dan Implementasi Median Filtering Untuk Peningkatan Akurasi Penghitung Push-up Real-Time Berbasis Deteksi Postur Tubuh

Sarwono, Wirandito (2026) Analisis Dan Implementasi Median Filtering Untuk Peningkatan Akurasi Penghitung Push-up Real-Time Berbasis Deteksi Postur Tubuh. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221088-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Tren kebugaran berbasis rumah yang berkembang pasca pandemi COVID-19 mendorong kebutuhan sistem pemantauan olahraga mandiri yang akurat dan terjangkau. Penelitian ini menganalisis dan mengimplementasikan teknik median filtering dalam dual-stage smoothing pipeline untuk meningkatkan akurasi sistem penghitung push-up real-time berbasis pose estimation pada perangkat mobile Android. Sistem dibangun di atas aplikasi Android fitness tracker menggunakan BlazePose melalui Google ML Kit sebagai model pose estimation. Pipeline yang diimplementasikan terdiri dari dua tahap: Stage 1 menggunakan Exponential Moving Average (EMA, alpha = 0,2) untuk meredam Gaussian noise yang terdistribusi kontinu, dan Stage 2 menggunakan median filter dengan window 5 frame dan threshold eliminasi outlier 20° untuk menangani impulsive noise berupa spike. Penghitungan repetisi dikendalikan oleh finite state machine dengan threshold asimetris 125°/110° dan timing protection 500 milidetik. Pengujian komparatif dilakukan dalam empat skenario dengan variasi kecepatan dan sudut kamera. Full Pipeline mencapai akurasi rata-rata 97,5%, lebih tinggi dari EMA saja (95,0%) dan tanpa filter (85,0%). Peningkatan paling signifikan terjadi pada skenario gerakan cepat, di mana Full Pipeline mencapai 100% berbanding 70% tanpa filter. Median filter berhasil mengeliminasi 110 dari 111 kejadian spike (99,1%), membuktikan efektivitasnya dalam menekan false count pada kondisi gerakan dinamis.
======================================================================================================================================
The growth of home-based fitness following the COVID-19 pandemic has increased demand for accurate, self-supervised exercise monitoring systems. This study analyzes and implements median filtering as part of a dual-stage smoothing pipeline to improve repetition counting accuracy in a real-time push-up counter running on an Android mobile device. The system is built on an Android fitness tracker application using BlazePose via Google ML Kit as the pose estimation backbone. The pipeline operates in two sequential stages: Stage 1 applies Exponential Moving Average (EMA, alpha = 0.2) to attenuate continuous Gaussian noise, while Stage 2 applies a 5-frame median filter with a 20° outlier rejection threshold to suppress impulsive spike noise. Repetition counting is governed by a finite state machine with asymmetric dual thresholds (125°/110°) and a 500 ms timing protection mechanism. Comparative testing across four scenarios with varying movement speeds and camera angles showed that the Full Pipeline achieved 97.5% mean accuracy, outperforming EMA-only (95.0%) and the unfiltered baseline (85.0%). The improvement was most pronounced under fast motion, where the Full Pipeline reached 100% accuracy versus 70% without filtering. The median filter eliminated 110 of 111 detected spike events (99.1%), confirming its effectiveness in reducing false counts under dynamic movement conditions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	median filtering, penghitung push-up, pose estimation, BlazePose, signal smoothing, finite state machine, exponential moving average, computer vision kebugaran, median filtering, push-up counter, pose estimation, BlazePose, signal smoothing, finite state machine, exponential moving average, fitness computer vision
Subjects:	Q Science > QA Mathematics > QA76.774.A53 Android Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Wirandito Sarwono
Date Deposited:	21 Jul 2026 07:09
Last Modified:	21 Jul 2026 07:09
URI:	http://repository.its.ac.id/id/eprint/135972

Actions (login required)

View Item