MULTI-MODAL LIGHTWEIGHT ELDERLY FALL RECOGNITION ALGORITHM FOR EMBEDDED EDGE HOME MONITORING DEVICES
Keywords:
MediaPipe pose, Fall detection, Multi-modal data fusion, Edge computing, Lightweight pose estimationAbstract
Machine vision-based elderly fall monitoring has become a mainstream solution for home elderly care, yet existing single-camera vision schemes suffer from low recognition accuracy for hunched and slow-moving seniors, frequent false alarms caused by illumination changes and furniture occlusion, and heavy computation load that fails real-time inference on low-cost embedded hardware such as Raspberry Pi. This paper develops a multimodal lightweight fall recognition framework based on MediaPipe Pose. A self-built dataset containing real home actions of elderly people is constructed, and transfer learning is adopted to fine-tune the pose model for special aged postures. Ultrasonic ranging, offline voice recognition and gas detection sensors are integrated to build weighted multi-modal joint discrimination logic. Dynamic frame extraction and dual-chip collaborative computing scheduling are designed to cut hardware computation overhead. The whole algorithm completes all inference locally without uploading video to cloud servers, effectively protecting household privacy. Experimental results verify that the proposed method suppresses false alarms significantly and supports stable real-time operation on low-cost edge terminals. It provides a feasible lightweight optimization scheme for privacy-sensitive home elderly monitoring systems.References
[1] Goyal N, Singh T, Goraya S M. Explainable Artificial Intelligence with Deep Convolutional Neural Networks for Real-Time Image-Based Yoga Posture Recognition in Fitness Training. New Generation Computing, 2025, 44(1): 5.
[2] Peng Wei, Deng Qiang, Li Ming, et al. Adaptive Multi‐Lens Phase Modulation for Scale‐Aware Privacy‐Preserving Human Pose Recognition. IET Software, 2025, 2025(1): 7879383.
[3] Song Ze, Li Gang, Song Ming. Adaptive Attention Fusion Method for Fine-Grained Human Posture Recognition Based on Micro-Doppler. Journal of Physics: Conference Series, 2025, 3166(1): 012010.
[4] Junaid I M, Madarapu S, Ari S. Human gait recognition using attention based convolutional network with sequential learning. Signal, Image and Video Processing, 2024, 19(1): 157.
[5] Ali N, Alec J, T L S, et al. Real world validation of activity recognition algorithm and development of novel behavioral biomarkers of falls in aged control and movement disorder patients. Frontiers in Aging Neuroscience, 2023, 15: 1117802.
[6] Liu Ziyang, Li Wei, Xing Gang, et al. The Identification of Elderly People with High Fall Risk Using Machine Learning Algorithms. Healthcare, 2022, 11(1): 47.
[7] A A M A, M A H, Abdelghani D, et al. The Applications of Metaheuristics for Human Activity Recognition and Fall Detection Using Wearable Sensors: A Comprehensive Analysis. Biosensors, 2022, 12(10): 821.
[8] Amandine D, Titus B, JeanPierre B. Identifying Fall Risk Predictors by Monitoring Daily Activities at Home Using a Depth Sensor Coupled to Machine Learning Algorithms. Sensors, 2021, 21(6): 1957.
[9] Huang Kun, Yuan Qiongqian, Huang Zefan. A Two-Stage Fall Recognition Algorithm Based on Human Posture Features. Sensors (Basel, Switzerland), 2020, 20(23).
[10] Ye Cheng, Li Jie, Hao Sen, et al. Identification of elders at higher risk for fall with statewide electronic health records and a machine learning algorithm. International Journal of Medical Informatics, 2020, 137: 104105.