Klasifikasi Postur Pemanasan: Pendekatan Deep Learning Terinspirasi oleh MediaPipe untuk Aplikasi Mobile

Mulyono, Luky Anggana (2025) Klasifikasi Postur Pemanasan: Pendekatan Deep Learning Terinspirasi oleh MediaPipe untuk Aplikasi Mobile. Other thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 5024201026-Undergraduate_Thesis.pdf] Text
5024201026-Undergraduate_Thesis.pdf - Accepted Version
Restricted to Repository staff only until 1 April 2027.

Download (2MB) | Request a copy

Abstract

Pemanasan dan peregangan tubuh sebelum olahraga sangat penting untuk meningkatkan fleksibilitas, mencegah cedera, dan memaksimalkan performa. Namun, banyak orang kesulitan memahami teknik pemanasan yang benar dan optimal tanpa bantuan pelatih. Penelitian ini mengembangkan aplikasi pendeteksi postur tubuh menggunakan framework MediaPipe untuk membantu pengguna mempraktekkan pemanasan secara mandiri. MediaPipe digunakan untuk mendeteksi landmark tubuh, memanfaatkan model BlazePose berbasis Convolutional Neural Network (CNN) untuk deteksi dan estimasi postur, serta algoritma K-Nearest Neighbor (K-NN) untuk klasifikasi postur. Dataset postur terdiri dari 10 tipe pemanasan statis yang direkam dalam berbagai sudut. Setelah dataset diproses menggunakan bantu MediaPipe, aplikasi dikembangkan di Android Studio dengan integrasi Google ML Kit Pose Detection API untuk deteksi dan klasifikasi secara real-time. Pengujian dilakukan pada 4 sudut dalam kondisi pencahayaan yang cukup, dengan akurasi rata-rata sebesar 90 \%. Sudut depan menghasilkan tingkat deteksi terbaik, sedangkan sudut belakang menunjukkan penurunan akurasi akibat adanya halangan dan overlap. Aplikasi ini memberikan umpan balik berupa nama postur yang diperagakan, tingkat confidence, dan countdown untuk memastikan durasi pemanasan yang tepat. Penelitian ini membuktikan efektivitas teknologi untuk meningkatkan pengalaman pemanasan yang lebih aman dan efisien.
==============================================================================================================================
Warm-up and body stretching before exercising are essential for improving flexibility, preventing injuries, and maximizing performance. However, many people struggle to understand proper and optimal warm-up techniques without a coach's guidance. This research developed a body posture detection application using the MediaPipe framework to help users perform warm-ups independently. MediaPipe was used to detect body landmarks, utilizing the BlazePose model based on Convolutional Neural Networks (CNN) for posture detection and estimation, as well as the K-Nearest Neighbor (K-NN) algorithm for posture classification.The posture dataset consisted of 10 types of static warm-ups recorded from various angles. After processing the dataset using MediaPipe tools, the application was developed in Android Studio with the integration of the Google ML Kit Pose Detection API for real-time detection and classification. Testing was conducted at four angles under adequate lighting conditions, achieving an average accuracy of 90\%. The front angle yielded the best detection rate, while the rear angle showed a decrease in accuracy due to obstructions and overlaps.The application provides feedback in the form of the name of the demonstrated posture, confidence levels, and a countdown timer to ensure proper warm-up duration. This research demonstrates the effectiveness of technology in enhancing a safer and more efficient warm-up experience.

Item Type: Thesis (Other)
Uncontrolled Keywords: Application, MediaPipe, ML Kit, Posture Classification, Stretching, Aplikasi, Klasifikasi Postur, Pemanasan
Subjects: T Technology > T Technology (General) > T58.5 Information technology. IT--Auditing
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User: Luky Anggana Mulyono
Date Deposited: 01 Feb 2025 23:50
Last Modified: 01 Feb 2025 23:50
URI: http://repository.its.ac.id/id/eprint/116486

Actions (login required)

View Item View Item