Klasifikasi Pose Jatuh Pada Lansia Berbasis Lidar Menggunakan Metode Machine Learning

Miawarni, Herti (2026) Klasifikasi Pose Jatuh Pada Lansia Berbasis Lidar Menggunakan Metode Machine Learning. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 07111960010010-Doctoral.pdf - Accepted Version Restricted to Repository staff only Download (47MB) | Request a copy

Abstract

Insiden jatuh pada lansia (lanjut usia) saat beraktivitas sehari-hari dapat menimbulkan gangguan kesehatan serius, menjadikannya salah satu risiko terbesar terutama bagi lansia yang tinggal sendiri. Untuk mendukung kehidupan mandiri serta mengurangi beban biaya perawatan, berbagai teknologi dikembangkan, termasuk Fall Detection System (FDS). Teknologi ini bertujuan mendeteksi insiden jatuh secara otomatis dan segera mengirimkan notifikasi ke keluarga atau tenaga medis. FDS melibatkan berbagai aspek
seperti Internet of Things, Wireless Sensor Network, teknologi sensor, serta peningkatan kinerja sistem melalui algoritma machine learning dan deep learning. Berdasarkan jenis sensor, FDS terbagi menjadi tiga kelompok: berbasis kamera, wearable, dan ambient. Kamera menawarkan informasi visual yang kaya, tetapi menimbulkan masalah privasi. Sensor wearable memberikan data akurat karena menempel langsung pada tubuh, namun mengganggu kenyamanan. Sensor ambient lebih ramah privasi dan nyaman, tetapi kompleks dalam pemasangan serta rentan terhadap kondisi lingkungan. Penelitian ini mengusulkan FDS berbasis sensor LiDAR sebagai bagian dari kategori ambient, karena mampu menjaga privasi, tidak membatasi gerak, mudah dipasang, serta tahan terhadap gangguan cahaya atau suara. Tujuan penelitian adalah merealisasikan klasifikasi pose jatuh pada FDS berbasis LiDAR yang efisien dan akurat dalam mendeteksi jatuh, sekaligus mengevaluasi keunggulan LiDAR dibanding pendekatan lain. Penelitian ini juga mengembangkan pendekatan representasi data baru dengan mengonversi data 2D-LiDAR menjadi citra statis untuk klasifikasi menggunakan algoritma
K-NN, RF, dan SVM. Selain itu, dilakukan modifikasi perangkat menjadi lowcost 3D-LiDAR dan pengembangan metode klasifikasi pose jatuh pada point cloud 3D-LiDAR melalui usulan PointLARKA. Penelitian ini dibatasi pada skenario indoor statis, dengan objek satu lansia dan sensor tunggal. Pada 2D-LiDAR, difokuskan pada akuisisi data dan penerapan algoritma machine learning untuk klasifikasi pose jatuh. Sedangkan pada 3D-LiDAR, fokus pada usulan metode klasifikasi pose jatuh. Evaluasi dilakukan terhadap akurasi, presisi, sensitivitas, dan parameter performansi lainnya. Hasil menunjukkan bahwa klasifikasi pose jatuh berbasis LiDAR efektif dalam membedakan aktivitas normal dan jatuh, dengan akurasi tinggi baik pada model 2D maupun 3D. PointLARKA secara khusus menunjukkan performa unggul dalam menangkap fitur spasial tubuh manusia yang kompleks dan tidak kaku, menjadikannya pendekatan yang menjanjikan untuk klasifikasi pose jatuh berbasis sensor non-invasif.
=====================================================================================================================================
Fall incidents among the elderly during daily activities can lead to serious health complications, making them one of the most significant risks, particularly for older adults living alone. To support independent living and reduce healthcare costs, various technologies have been developed, including the Fall Detection System (FDS). This technology aims to automatically detect fall incidents and promptly notify family members or medical personnel. FDS integrates various components, such as the Internet of Things (IoT), Wireless Sensor Networks (WSN), sensor technologies, and system performance optimization using machine learning and deep learning algorithms. Based on sensing modalities, FDS is categorized into three types: camerabased, wearable-based, and ambient-based systems. Camera-based systems provide rich visual data but raise privacy concerns. Wearable sensors offer high accuracy due to direct contact with the body but may cause discomfort. Ambient sensors, while more privacy-preserving and comfortable, tend to be complex to install and are sensitive to environmental disturbances. This study proposes a LiDAR-based FDS, categorized as ambient, which ensures user privacy, allows free movement, is easy to install, and is resilient to lighting and noise disturbances. The objective of this study is to realize an efficient and accurate LiDAR-based Fall Detection System (FDS) for fall-pose classification, while also evaluating the advantages of LiDAR compared with other approaches. This research develops a new data representation by converting 2D-LiDAR measurements into static images for classification using k-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Machine (SVM). In addition, the hardware is modified into a low-cost 3D-LiDAR setup, and a 3D-LiDAR point-cloud–based fall-pose classification method is developed through the proposed PointLARKA. This study is limited to a static indoor scenario with a single elderly subject and a single sensor. For 2D-LiDAR, the focus is on data acquisition and the application of machine learning algorithms for fallpose classification. For 3D-LiDAR, the focus is on the proposed fall-pose classification method. Performance is evaluated using accuracy, precision, sensitivity, and other relevant metrics. The results show that LiDAR-based fall-pose classification effectively distinguishes normal activities from falls, achieving high accuracy for both 2D and 3D models. PointLARKA, in particular, demonstrates strong performance in capturing complex and non-rigid spatial features of the human body, making it a promising non-invasive sensing approach for fall-pose classification.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Ambient-based, Fall Detection System, Fall-pose classification, LiDAR, Machine Learning
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK6592.O6 Optical radar T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20001-(S3) PhD Thesis
Depositing User:	Herti Miawarni
Date Deposited:	28 Jan 2026 03:52
Last Modified:	28 Jan 2026 03:52
URI:	http://repository.its.ac.id/id/eprint/130777

Actions (login required)

View Item