A lightweight fall-detection device integrating ToF laser ranging and visual sensing
Liangli Wang
Guilin University of Electronic Technology
Yulin Song
Guilin University of Electronic Technology
Jinfei Nong
Guilin University of Electronic Technology
Kai Guo
Guilin University of Electronic Technology
Xuan Pan
Guilin University of Electronic Technology
DOI: https://doi.org/10.59429/esta.v13i3.15562
Keywords: ToF laser ranging; visual sensing; fall detection; multimodal fusion; edge intelligence; simulation verification
Abstract
To address the instability of vision-only fall detection under low-light conditions and occlusion, as well as the constraints of edge computing, we designed a lightweight device that integrates multi-zone time-of-flight (ToF) laser ranging with low-resolution visual sensing and established a reproducible verification workflow based on digital human simulation. The ToF branch extracts distance, vertical position, descent velocity, effective ranging ratio, and near-ground region ratio, while the visual branch extracts the human body bounding box, centroid, tilt angle, motion amplitude, detection confidence, and brightness. Features from the two branches are gated and fused according to real-time quality scores, after which a depthwise separable temporal convolution identifies the continuous upright-descent-near-ground progression. The simulator generated 9,600 sequences across 32 virtual body profiles, including 3,840 fall sequences and 5,760 daily-activity sequences, under low light, backlighting, occlusion, invalid ToF echoes, and structured unimodal distortions. In composite stress tests using splits that were disjoint by virtual body profile, the proposed method achieved accuracy, precision, recall, and F1 scores of 96.89%, 96.89%, 95.28%, and 96.08%, respectively, with 1.22 false alarms per hour and an average alarm latency of 88.7 ms. The model contains 14,532 parameters and 1.42 M MACs; under the equivalent-cost model for a quadcore ARM processor, the INT8 end-to-end latency was 12.8 ms. The results indicate that quality gating can mitigate the performance degradation caused by unimodal disturbances, although these findings still require validation using physical prototypes and real-world scenarios.
References
[1] Mohan, D., Al-Hamid, D. Z., Chong, P. H. J., Sudheera, K. L. K., Gutierrez, J., Chan, H. C. B., & Li, H. (2024). Artificial intelligence and IoT in elderly fall prevention: A review. IEEE Sensors Journal, 24(4), 4181–4198. https://doi.org/10.1109/JSEN.2023.3344605.
[2] Alam, E., Sufian, A., Dutta, P., & Leo, M. (2022). Vision-based human fall detection systems using deep learning: A review. Computers in Biology and Medicine, 146, 105626. https://doi.org/10.1016/j.compbiomed.2022.105626.
[3] Gaya-Morey, F. X., Manresa-Yee, C., & Buades-Rubio, J. M. (2024). Deep learning for computer vision based activity recognition and fall detection of the elderly: A systematic review. Applied Intelligence, 54(19), 8982–9007. https://doi.org/10.1007/s10489-024-05645-1.
[4] Qi, P., Chiaro, D., & Piccialli, F. (2023). FL-FD: Federated learning-based fall detection with multimodal data fusion. Information Fusion, 99, 101890. https://doi.org/10.1016/j.inffus.2023.101890.
[5] Amsaprabhaa, M., Jane, Y. N., & Nehemiah, H. K. (2023). Multimodal spatiotemporal skeletal kinematic gait feature fusion for vision-based fall detection. Expert Systems with Applications, 212, 118681. https://doi.org/10.1016/j.eswa.2022.118681.
[6] Jiao, S., Li, G., Zhang, G., Zhou, J., & Li, J. (2024). Multimodal fall detection for solitary individuals based on audio-video decision fusion processing. Heliyon, 10(8), e29596. https://doi.org/10.1016/j.heliyon.2024.e29596.
[7] Lafuente-Arroyo, S., Martín-Martín, P., Iglesias-Iglesias, C., Maldonado-Bascón, S., & Acevedo-Rodríguez, F. J. (2022). RGB camera-based fallen person detection system embedded on a mobile platform. Expert Systems with Applications, 197, 116715. https://doi.org/10.1016/j.eswa.2022.116715.
[8] Lin, B.-S., Yu, T., Peng, C.-W., Lin, C.-H., Hsu, H.-K., Lee, I.-J., & Zhang, Z. (2022). Fall detection system with artificial intelligence-based edge computing. IEEE Access, 10, 4328–4339. https://doi.org/10.1109/ACCESS.2021.3140164.
[9] Romeo, L., Marani, R., D’Orazio, T., & Cicirelli, G. (2023). Video based mobility monitoring of elderly people using deep learning models. IEEE Access, 11, 2804–2819. https://doi.org/10.1109/ACCESS.2023.3234421.