Implementasi ML Kit Pose Landmarking pada Aplikasi Mobile Penghitung Push-Up Berbasis Android

Zafir, Muhammad Luthfi (2026) Implementasi ML Kit Pose Landmarking pada Aplikasi Mobile Penghitung Push-Up Berbasis Android. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221095-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

Latihan push-up banyak digunakan dalam rutinitas kebugaran, namun melakukannya secara mandiri menimbulkan tantangan dalam memantau kualitas gerakan dan menghitung repetisi secara akurat. Penelitian ini mengimplementasikan ML Kit Pose Landmarking dalam aplikasi push-up berbasis Android untuk mengatasi tantangan tersebut melalui computer vision yang berjalan secara on-device. Sistem dikembangkan menggunakan model Software Development Life Cycle (SDLC) Prototyping, diawali dengan proof-of-concept pada lingkungan desktop menggunakan Python dan RTMPose, kemudian diimplementasikan penuh pada Android menggunakan Kotlin, Jetpack Compose, CameraX, dan Google ML Kit Pose Detection dalam mode accurate dengan STREAM_MODE.
Implementasi mencakup tiga aspek: (1) deteksi real-time terhadap 33 titik landmark tubuh menggunakan AccuratePoseDetectorOptions, dengan seleksi sisi tubuh otomatis berdasarkan nilai inFrameLikelihood; (2) visualisasi skeleton overlay yang dirender melalui Canvas Jetpack Compose yang ditumpangkan di atas live preview kamera menggunakan transformasi koordinat fill-center; serta (3) penghitungan repetisi otomatis menggunakan Finite State Machine dua state ("up" dan "down") yang sepenuhnya digerakkan oleh koordinat landmark.
Pengujian pada Samsung Galaxy Z Flip 3 menunjukkan akurasi deteksi landmark sebesar 100% pada seluruh kondisi pencahayaan dan latar belakang yang diujikan. Sistem mencapai frame rate efektif ~30fps dengan rata-rata latensi deteksi ~3ms per frame dan konsumsi RAM ~170MB. Seluruh enam skenario pengujian fungsional black-box berhasil dilewati, termasuk operasi tanpa koneksi internet. Hasil ini mengonfirmasi bahwa ML Kit Pose Landmarking menyediakan fondasi yang andal dan efisien untuk deteksi push-up secara on-device pada platform Android.
==================================================================================================================================
Push-up exercises are widely used in fitness routines, yet performing them independently poses challenges in monitoring movement quality and accurately counting repetitions. This research implements ML Kit Pose Landmarking in an Android-based push-up application to address these challenges through on-device computer vision. The system was developed using the Software Development Life Cycle (SDLC) Prototyping model, beginning with a desktop proof-of-concept using Python and RTMPose, followed by full implementation on Android using Kotlin, Jetpack Compose, CameraX, and Google ML Kit Pose Detection in accurate mode with STREAM_MODE.
The implementation covers three aspects: (1) real-time detection of 33 body landmarks using AccuratePoseDetectorOptions, with automatic body side selection based on inFrameLikelihood values; (2) visualization of a skeleton overlay rendered via Jetpack Compose Canvas overlaid on the live camera preview using a fill-center coordinate transformation; and (3) automatic repetition counting using a two-state Finite State Machine ("up" and "down") driven entirely by landmark coordinates.
Testing on a Samsung Galaxy Z Flip 3 demonstrated 100% landmark detection accuracy across all lighting and background conditions. The system achieved an effective frame rate of ~30fps with an average detection latency of ~3ms per frame and ~170MB RAM consumption. All six functional black-box test scenarios passed, including offline operation. These results confirm that ML Kit Pose Landmarking provides a reliable and efficient foundation for on-device push-up detection on Android.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	ML Kit, pose estimation, landmark detection, push-up counter, Android, Jetpack Compose, CameraX, BlazePose, ML Kit, pose estimation, landmark detection, push-up counter, Android, Jetpack Compose, CameraX, BlazePose
Subjects:	T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Luthfi Zafir
Date Deposited:	24 Jul 2026 04:06
Last Modified:	24 Jul 2026 04:07
URI:	http://repository.its.ac.id/id/eprint/136860

Actions (login required)

View Item