Overall, the study demonstrates that combining robust biomechanical features with multimodal sensing and deployment-aware resource management provides an effective foundation for practical privacy-preserving fall detection.
Abstract
Falls among older adults can lead to severe injury when assistance is delayed. Vision-based monitoring offers a contactless means of detecting falls, but deploying such systems in indoor environments raises important privacy and implementation challenges. This study presents a privacy-aware fall detection framework that combines pose-derived biomechanical features, wearable inertial sensing, and deployment-aware computation within an edge–fog–cloud architecture. The proposed approach is evaluated on three public datasets (UR Fall, UP-Fall, and Le2i). Static biomechanical features achieve an F1-score of 99.95% on UR Fall and above 91% on Le2i, whereas performance on the more challenging UP-Fall dataset reaches approximately 80% F1-score. The effect of temporal aggregation is found to depend on dataset characteristics, with larger temporal windows improving performance in some cases while providing little benefit in others. Cross-dataset evaluation reveals substantial performance degradation under dataset shift, indicating limited generalization across different recording conditions and participant populations. To investigate this behaviour, feature distributions are analysed using Jensen–Shannon divergence and Wasserstein distance, and robustness-oriented feature selection improves transfer performance compared with using the complete temporal feature set. Multimodal fusion of pose-based predictions with inertial sensing further improves recall and F1-score on the UP-Fall dataset. End-to-end deployment profiling on a Raspberry Pi 5 demonstrates that the complete inference pipeline, including pose estimation, biomechanical feature extraction, temporal feature aggregation, and classification, operates in real time on resource-constrained hardware. Deployment analysis using iFogSim further shows that privacy-aware placement strategies substantially reduce network communication while providing favourable trade-offs among latency, privacy, and resource utilization compared with cloud-centric execution. Overall, the study demonstrates that combining robust biomechanical features with multimodal sensing and deployment-aware resource management provides an effective foundation for practical privacy-preserving fall detection.
Falls among older adults are a major safety challenge, but continuous monitoring is difficult to sustain. Video captures fall-related posture and motion, yet deployment is limited by privacy, computation, and bandwidth. Supervised pose estimation is anatomically interpretable but vulnerable to occlusion and partial body visibility. We propose a privacy-preserving framework that replaces RGB transmission with compact motion representations based on unsupervised keypoints and predictive temporal modeling. Local processing performs segmentation and keypoint extraction; variational recurrent prediction and sequence classification then detect falls from observed and forecasted motion. We evaluate the framework on the UR Fall Detection and Human Fall datasets using random, subject-disjoint, and occlusion-based splits. Under random splits, neither representation consistently dominates, suggesting that standard protocols may hide meaningful differences. Under subject-disjoint evaluation, supervised keypoints show a statistically significant advantage, but performance varies by subject: they perform better when anatomical landmarks are visible, whereas unsupervised keypoints are more robust to occlusion and partial visibility, though they produce more false positives for complex activities. Under occlusion-based evaluation, supervised keypoints miss nearly half of all falls, while unsupervised keypoints retain strong sensitivity and substantially outperform them. Their anatomical independence allows spatial anchors to adapt to visible body structure rather than fail on absent landmarks. The gap widens under bandwidth constraints, where supervised localization errors compound through the temporal model. These findings show that representation choice should reflect expected visual conditions and that unsupervised keypoints offer an advantage when body visibility is compromised.
Tasmiah Haque, Jacob Kosinski, S. Mohan et al.· arXiv.org· 0 citations
Accidental falls among the elderly demand highly reliable detection systems; however, existing solutions based on wearables, cameras, or WiFi often suffer from environmental interference, privacy concerns, and poor performance in detecting slow-onset falls (e.g., fainting). Despite these challenges, millimeter-wave (mmWave) radar-based methods have emerged as an attractive alternative. In this article, we propose an effective and efficient fall detection method, mmFallBbox, which leverages mmWave radar to track 3-D human bounding box dynamics. By exploiting the correlation between fall states and bounding box evolution, our approach effectively distinguishes between normal activities and slow fall states, such as fainting or medical conditions, which traditional systems struggle to detect. To evaluate the performance of mmFallBbox, we collected a large-scale fall detection dataset consisting of 60 h of radar data synchronized with video annotations. This dataset will be made publicly available to the research community for further development. We achieved an $F1$ score of 0.977 on our dataset and achieved state-of-the-art performance with limited computational complexity. Moreover, extensive experimental results show significant improvements in detecting slow-onset falls and providing explainable outputs, offering a promising solution for real-world fall detection applications.
Wenxuan Li, Dongheng Zhang, Jianwen Tong et al.· IEEE Transactions on Radar S...· 0 citations
The aging population has elevated falls into a critical public health issue. While camera-based YOLO algorithms offer non-contact detection, standard YOLOv13 struggles with occlusion, similar postures, and high computational demands. To address this, we propose RDD-YOLO, a task-oriented architecture optimized for fall detection accuracy and efficiency. Rather than simply stacking existing modules, RDD-YOLO assigns RepViTBlock to backbone feature extraction, DySample to detail-preserving feature fusion, and DHead to multi-scale regression, so that each component plays a complementary role in the detection pipeline. Evaluated on refined URFD and MCF datasets, RDD-YOLO outperformed YOLOv13, RT-DETR, and Faster-RCNN. For the URFD and MCF datasets, the mAP reached 92.1% and 87.3%, respectively. Speed tests showed that on a laptop (in a WSL2 environment), inference speed increased from approximately 47 FPS to approximately 57 FPS. Through pruning and FP16 half-precision inference, the speed further increased to over 70 FPS, demonstrating the feasibility of this method in real-time, resource-constrained application scenarios.
Falls among older adults are a leading cause of injury and hospitalisation, yet existing fall detection devices, such as wearable sensors or video monitoring, encounter issues due to user compliance and privacy concerns. This paper presents a low-cost, privacy preserving fall detection system using a rotating LiDAR sensor mounted on a servo which is controlled by a simple microcontroller. A full sweep of a room captures approximately ten thousand depth points, which are streamed to a PC for pre-processing, feature extraction, and classification. Key geometric and distributional features are derived from point clouds and classified using a Random Forest model. Evaluation on fifty-three labelled scans achieved ninety-six-point two percent overall accuracy with only two misclassifications. These results demonstrate that affordable LiDAR combined with lightweight machine learning offers a viable path toward privacy-preserving, fall detection in care environments.
Gary Alexander Warner· E3S Web of Conferences· 0 citations
The findings show that CNN-based architectures dominate algorithm choice, edge devices dominate deployment platforms, and optimization remains central to real-time inference on constrained hardware.
Mahammad Nabizade, Réda Yahiaoui, Isabelle Lajoie et al.· Italian National Conference...· 0 citations
