Jatuh pada lansia merupakan risiko kesehatan yang serius, sering menyebabkan cedera, hilangnya kemandirian, dan peningkatan biaya perawatan kesehatan. Sistem pemantauan wearable muncul sebagai pendekatan praktis untuk mengatasi tantangan ini, menawarkan pemantauan real-time secara terus-menerus tanpa masalah privasi. Dengan menggunakan perangkat kecil dan nyaman yang mengukur gerakan tubuh, sistem ini dapat mendeteksi jatuh. Penelitian ini mengeksplorasi pengembangan solusi pemantauan wearable yang andal, mudah diakses, dan hemat biaya untuk mendukung perawatan lansia, menekankan peran teknologi dalam mendorong penuaan yang lebih aman.
Penerapan deep learning pada perangkat dengan sumber daya terbatas masih jarang diteliti, terutama untuk deteksi jatuh pada lansia secara real-time. Studi ini mengembangkan model, menggunakan mikrokontroler ESP32-S3 dan sensor IMU MPU6050, memanfaatkan CNN yang dilatih pada dataset SisFall dan KFall. Pra-pemrosesan mencakup filter moving average untuk mengurangi noise dan feature scaling untuk konsistensi training data. Model kemudian di-kuantisasi ke format 8-bit integer (int8) menggunakan TensorFlow Lite, meminimalkan kebutuhan memori dan komputasi untuk deployment. Hasilnya mencapai akurasi sekitar 97?ngan waktu inferensi rata-rata 4,9 milidetik. Temuan ini menunjukkan kelayakan penerapan deep learning pada mikrokontroler untuk aplikasi pemantauan kesehatan real-time.
Falls among the elderly pose a serious health risk, often leading to injuries, loss of independence, and increased healthcare costs. Wearable monitoring systems have emerged as a practical approach to address this challenge, offering continuous real-time monitoring without privacy concerns. By using small and comfortable devices that track body movements, these systems can detect falls. This study explores the development of a reliable, accessible, and cost-effective wearable monitoring solution to support elderly care, highlighting the role of technology in promoting safer aging.
The application of deep learning on resource-constrained devices remains underexplored, especially for real-time fall detection in the elderly. This study develops a model using the ESP32-S3 microcontroller and the MPU6050 IMU sensor, leveraging a CNN trained on the SisFall and KFall datasets. Preprocessing includes a moving average filter to reduce noise and feature scaling for data consistency during training. The model is then quantized into 8-bit integer (int8) format using TensorFlow Lite, minimizing memory and computational requirements for deployment. The results achieve approximately 97?curacy with an average inference time of 4.9 milliseconds. These findings demonstrate the feasibility of deploying deep learning on microcontrollers for real-time health monitoring applications.
<button class="text-token-text-secondary hover:bg-token-bg-secondary rounded-lg" aria-label="Copy" aria-selected="false" data-testid="copy-turn-action-button" data-state="closed"><svg width="20" height="20" viewBox="0 0 20 20" fill="currentColor" xmlns="http://www.w3.org/2000/svg" class="icon"></svg></button><button class="text-token-text-secondary hover:bg-token-bg-secondary rounded-lg" aria-label="Good response" aria-selected="false" data-testid="good-response-turn-action-button" data-state="closed"><svg width="20" height="20" viewBox="0 0 20 20" fill="currentColor" xmlns="http://www.w3.org/2000/svg" class="icon"></svg></button><button class="text-token-text-secondary hover:bg-token-bg-secondary rounded-lg" aria-label="Bad response" aria-selected="false" data-testid="bad-response-turn-action-button" data-state="closed"><svg width="20" height="20" viewBox="0 0 20 20" fill="currentColor" xmlns="http://www.w3.org/2000/svg" class="icon"></svg></button><button class="text-token-text-secondary hover:bg-token-bg-secondary rounded-lg" aria-label="Read aloud" aria-selected="false" data-testid="voice-play-turn-action-button" data-state="closed"><svg width="20" height="20" viewBox="0 0 20 20" fill="currentColor" xmlns="http://www.w3.org/2000/svg" class="icon"></svg></button><button class="text-token-text-secondary hover:bg-token-bg-secondary rounded-lg" aria-label="Edit in canvas" aria-selected="false" data-state="closed"><svg width="20" height="20" viewBox="0 0 20 20" fill="currentColor" xmlns="http://www.w3.org/2000/svg" class="icon"></svg></button><button class="text-token-text-secondary hover:bg-token-bg-secondary rounded-lg" aria-label="Share" aria-selected="false" data-state="closed"><svg width="20" height="20" viewBox="0 0 20 20" fill="currentColor" xmlns="http://www.w3.org/2000/svg" class="icon"></svg></button><button class="cursor-pointer text-token-text-secondary hover:bg-token-bg-secondary touch:px-2.5 h-[30px] rounded-md px-1.5" aria-label="Switch model" type="button" id="radix-«r4r»" aria-haspopup="menu" aria-expanded="false" data-state="closed"><svg width="20" height="20" viewBox="0 0 20 20" fill="currentColor" xmlns="http://www.w3.org/2000/svg" class="icon">
</svg><svg width="16" height="16" viewBox="0 0 16 16" fill="currentColor" xmlns="http://www.w3.org/2000/svg" class="icon-sm">
</svg>
<button class="btn relative btn-secondary shadow-long flex rounded-xl border-none active:opacity-1">
<svg width="20" height="20" viewBox="0 0 20 20" fill="currentColor" xmlns="http://www.w3.org/2000/svg" class="icon"></svg>Ask ChatGPT </button>
Kata Kunci : Kecerdasan Buatan, Pembelajaran Dalam, Monitoring Berkelanjutan, Sensor yang Digunakan, Pendeteksi Jatuh.
