EXPLAINABLE RULE-BASED FALL DETECTION FOR EDGE DEVICES USING MEDIAPIPE
DOI:
https://doi.org/10.54314/jssr.v9i4.6867Keywords:
Periodic Table of Elements, Chemistry Teaching Materials, Spiritual Values, Discovery LearningAbstract
Abstract: Falls represent a leading cause of injury-related mortality among elderly individuals, demanding reliable automated detection deployable in resource-constrained environments. Existing vision-based approaches predominantly employ machine learning classifiers requiring large labelled datasets, substantial training pipelines, and GPU-class hardware, impeding rapid clinical deployment. This paper presents an explainable, fully rule-based real-time fall detection system implemented in Python using MediaPipe Pose, designed to operate on commodity edge devices without any training dataset. The system comprises three tightly integrated layers: (1) a mathematical formulation layer computing vector-based torso angle, distance-ratio body normalization, and frame-to-frame CoM velocity; (2) a temporal modeling layer combining a deterministic four-state Finite State Machine with a sliding-window motion energy analysis; and (3) a robustness evaluation across three camera viewing angles, two lighting conditions, and three fall types. Experimental validation yields a sensitivity of 91.8%, specificity of 95.8%, and F1-score of 93.1%, with a mean end-to-end detection latency of 140.1 ms on Raspberry Pi 4—satisfying a 200 ms real-time constraint while consuming only 72.5% mean CPU and 284 MB RAM. The rule-based architecture provides full decision auditability without black-box models, aligning with emerging XAI requirements for medical device regulation.
Keywords: Periodic Table of Elements; Chemistry Teaching Materials; Spiritual Values; Discovery Learning.
Abstrak: Kondisi Jatuh merupakan penyebab utama kematian akibat cedera di kalangan lansia, sehingga membutuhkan deteksi otomatis yang andal dan dapat diterapkan di lingkungan dengan keterbatasan sumber daya. Pendekatan berbasis visi yang ada sebagian besar menggunakan pengklasifikasi pembelajaran mesin yang membutuhkan kumpulan data berlabel besar, alur pelatihan yang substansial, dan perangkat keras kelas GPU, yang menghambat penerapan klinis yang cepat. Makalah ini menyajikan sistem deteksi jatuh waktu nyata yang dapat dijelaskan dan sepenuhnya berbasis aturan yang diimplementasikan dalam Python menggunakan MediaPipe Pose, yang dirancang untuk beroperasi pada perangkat edge komoditas tanpa kumpulan data pelatihan apa pun. Sistem ini terdiri dari tiga lapisan yang terintegrasi erat: (1) lapisan formulasi matematika yang menghitung sudut tubuh berbasis vektor, normalisasi tubuh rasio jarak, dan kecepatan CoM antar frame; (2) lapisan pemodelan temporal yang menggabungkan Mesin Keadaan Terbatas empat keadaan deterministik dengan analisis energi gerak jendela geser; dan (3) evaluasi ketahanan di tiga sudut pandang kamera, dua kondisi pencahayaan, dan tiga jenis jatuh. Validasi eksperimental menghasilkan sensitivitas 91,8%, spesifisitas 95,8%, dan skor F1 93,1%, dengan latensi deteksi ujung-ke-ujung rata-rata 140,1 ms pada Raspberry Pi 4—memenuhi batasan waktu nyata 200 ms sambil hanya mengonsumsi 72,5% CPU rata-rata dan 284 MB RAM. Arsitektur berbasis aturan memberikan kemampuan audit keputusan penuh tanpa model kotak hitam, selaras dengan persyaratan XAI yang muncul untuk regulasi perangkat medis.
Kata Kunci: Deteksi jatuh, MediaPipe Pose, komputasi tepi, sistem berbasis aturan, kecerdasan buatan dengan keadaan terbatas, AI yang dapat dijelaskan, perawatan lansia, pemantauan real-time
Downloads
References
Al-Ayyoub, M. A., et al. (2024a). YOLOv8-based human fall detection with pose estimation. IEEE Access, 12, 14820–14835. https://doi.org/10.1109/ACCESS.2024.3358192
Al-Ayyoub, M. A., et al. (2024b). YOLOv8 for fall detection. IEEE Access, 12. https://doi.org/10.1109/ACCESS.2024.3358192
Alatiah, H. A., & Chelloug, S. A. (2023). Human action recognition using MediaPipe. Sensors, 23(2), 832. https://doi.org/10.3390/s23020832
Bazarevsky, V., et al. (2020). BlazePose: On-device real-time body pose tracking. arXiv. https://arxiv.org/abs/2006.10204
Bououden, B., et al. (2024). FSM-based control for embedded real-time monitoring in healthcare IoT. IEEE Transactions on Consumer Electronics, 70(1), 892–901. https://doi.org/10.1109/TCE.2023.3318841
Cao, Z., et al. (2021). OpenPose: Realtime multi-person 2D pose estimation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(1), 172–186. https://doi.org/10.1109/TPAMI.2019.2929257
Casilari, E., Santoyo-Ramón, J. A., & Cano-García, J. M. (2017). Analysis of public datasets for wearable fall detection. Sensors, 17(7), 1513. https://doi.org/10.3390/s17071513
Dhelim, S., et al. (2023). A survey on brain-inspired intelligent agents. IEEE Transactions on Neural Networks and Learning Systems, 34(10), 7662–7678. https://doi.org/10.1109/TNNLS.2022.3167517
European Parliament. (2024). Regulation (EU) 2024/1689 — AI Act. Official Journal of the European Union.
Godfrey, A., et al. (2018). From A to Z: Wearable technology explained. Maturitas, 113, 40–47. https://doi.org/10.1016/j.maturitas.2018.04.012
Gunning, D., et al. (2019). XAI—Explainable artificial intelligence. Science Robotics, 4(37), Article eaay7120. https://doi.org/10.1126/scirobotics.aay7120
Holzinger, A., et al. (2019). Causability and explainability of AI in medicine. WIREs Data Mining and Knowledge Discovery, 9(4), Article e1312. https://doi.org/10.1002/widm.1312
Jain, S., Bhattacharya, M., & Khanna, V. (2023). Bidirectional LSTM for skeleton-based fall detection. Computers in Biology and Medicine, 165, Article 107451. https://doi.org/10.1016/j.compbiomed.2023.107451
Karlsson, M. K., Magnusson, C., von Schewelov, E., & Rosengren, B. E. (2023). Prevention of falls in the elderly: A review. Osteoporosis International, 34(3), 425–446. https://doi.org/10.1007/s00198-022-06647-2
Li, P., Jiang, Y., & Zhao, K. (2023). Lightweight rule-based fall detection using landmark angles. Biomedical Signal Processing and Control, 85, Article 104882. https://doi.org/10.1016/j.bspc.2023.104882
Liu, X., et al. (2023a). Attention-enhanced transformer for fall detection. IEEE Transactions on Instrumentation and Measurement, 72. https://doi.org/10.1109/TIM.2023.3264911
Liu, X., et al. (2023b). Attention-enhanced transformer network for real-time fall detection. IEEE Transactions on Instrumentation and Measurement, 72, 1–12. https://doi.org/10.1109/TIM.2023.3264911
Lord, S. R., Sherrington, C., Menz, H. B., & Close, J. C. T. (2022). Falls in older people (3rd ed.). Cambridge University Press.
Lugaresi, C., et al. (2019). MediaPipe: A framework for building perception pipelines. arXiv. https://arxiv.org/abs/1906.08172
Luo, H., Yang, J., & Wu, W. (2023). Edge-assisted real-time human activity recognition. IEEE Internet of Things Journal, 10(7), 6165–6177. https://doi.org/10.1109/JIOT.2022.3222738
Luque-Nieto, M., Martinez-de-Dios, F. J., & Ollero, A. (2023). Vision-based fall detection for elderly care: A systematic review. Sensors, 23(4), 2185. https://doi.org/10.3390/s23042185
Madeleine, P., Samani, C., & Holtermann, A. (2023). Mediolateral CoP displacement during standing: Systematic review. Gait & Posture, 99, 231–241. https://doi.org/10.1016/j.gaitpost.2022.11.008
Menz, H. B., Lord, S. R., & Fitzpatrick, R. C. (2003). Age-related changes in walking stability. Age and Ageing, 32(2), 137–142. https://doi.org/10.1093/ageing/32.2.137
Merenda, M., Porcaro, C., & Iero, D. (2020). Edge machine learning for AI-enabled IoT. Sensors, 20(9), 2533. https://doi.org/10.3390/s20092533
Mohammadi Ghalehney, M., et al. (2023). Fall detection using posture angle and HMM. Expert Systems with Applications, 215, Article 119362. https://doi.org/10.1016/j.eswa.2022.119362
Molnar, C. (2022). Interpretable machine learning (2nd ed.). Leanpub. https://christophm.github.io/interpretable-ml-book
Montero-Odasso, A., et al. (2022). World guidelines for falls prevention and management for older adults. Age and Ageing, 51(9), Article afac205. https://doi.org/10.1093/ageing/afac205
Ren, L., & Meng, Y. (2023). Geometrical feature-based fall detection with posture analysis. Healthcare Technology Letters, 10(1–2), 15–22. https://doi.org/10.1049/htl2.12040
Roggen, D., et al. (2010). Collecting complex activity datasets in networked sensor environments. Proceedings of the IEEE International Conference on Networked Sensing Systems, 233–240.
Rundo, R., Ciaccio, G., & Vitabile, S. (2023). Explainable AI for healthcare. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 31, 2210–2220. https://doi.org/10.1109/TNSRE.2023.3268412
Sanchez-Iborra, R., & Skarmeta, A. F. (2021). TinyML-enabled frugal smart objects. IEEE Circuits and Systems Magazine, 21(2), 4–18. https://doi.org/10.1109/MCAS.2021.3051622
Siqueira, H., Kobs, K., & Berretti, S. (2023). Two-stream CNN for video-based fall detection. Pattern Recognition Letters, 176, 100–107. https://doi.org/10.1016/j.patrec.2023.11.002
Todd, C., Skelton, A., & Lafortune, S. (2024). Older people and falls. BMC Geriatrics, 24(1), 1–18. https://doi.org/10.1186/s12877-024-04820-6
Wang, F., et al. (2023). A comprehensive survey on fall detection systems for the elderly. IEEE Sensors Journal, 23(10), 10472–10491. https://doi.org/10.1109/JSEN.2023.3258874
Wang, Y., Li, H., & Chen, Z. (2023). Benchmarking pose estimation on edge platforms. IEEE Transactions on Industrial Informatics, 19(10), 10541–10551. https://doi.org/10.1109/TII.2023.3245671
Winter, D. A. (2009). Biomechanics and motor control of human movement (4th ed.). Wiley.
World Health Organization. (2023). Falls. WHO Fact Sheet. https://www.who.int/news-room/fact-sheets/detail/falls
Yadav, A., & Jha, C. K. (2023). Fall detection using skeleton-based graph convolutional network: A review. IEEE Access, 11, 32104–32124. https://doi.org/10.1109/ACCESS.2023.3262045
Zeng, L., Wang, Q., & Zhang, Y. (2023a). Spatial-temporal GCN for elderly fall detection. IEEE Journal of Biomedical and Health Informatics, 27(3), 1418–1427. https://doi.org/10.1109/JBHI.2022.3224310
Zeng, L., et al. (2023b). Spatial-temporal GCN for fall detection from RGB. IEEE Journal of Biomedical and Health Informatics, 27(3), 1418–1427. https://doi.org/10.1109/JBHI.2022.3224310
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Louis Ferdinand Boesday, Yetursance Y. Manafe, Tirsa J. Saruan, Abdi K. Radja (Author)
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
