Academic Journal of Computing & Information Science, 2024, 7(11); doi: 10.25236/AJCIS.2024.071111.
Shan Li1, Lei Ding2, Yi Shi1,3
1Electronic Science and Technology, Shaanxi University of Science and Technology, Xi'an, China
2Computer Science and Technology, Xi'an Jiaotong University, Xi'an, China
3Computer Science and Technology, Shaanxi University of Science and Technology, Xi'an, China
Falls are sudden accidental injuries, and a real-time fall detection system can mitigate the severe consequences of delayed detection by providing timely assistance to the individual who has fallen. In recent years, with the rapid development of technologies such as deep learning, various fall detection methods have been developed. Based on this, this paper proposes a multi-modal fall detection system that integrates accelerometers and video surveillance. The system simultaneously detects the target using both video surveillance and accelerometers, and then performs decision fusion to obtain the final detection result. Experimental evaluations on the UR Fall Detection dataset demonstrate that the developed multi-sensor fusion fall detection system achieves higher accuracy compared to single modality fall detection methods using either video surveillance or accelerometer sensors.
YOLOv5; LSTM; Decision Fusion; Multi-modal; Fall Detection
Shan Li, Lei Ding, Yi Shi. Research on Fall Detection Based on Vision and Wearable Devices. Academic Journal of Computing & Information Science (2024), Vol. 7, Issue 11: 78-85. https://doi.org/10.25236/AJCIS.2024.071111.
[1] World Health Organization. Global Health Estimates: Life expectancy and leading causes of death and disability[EB/OL]. Retrieved from https://www.who.int/data/gho,2024.
[2] SUCERQUIA, Angela; LÓPEZ, José David; VARGAS-BONILLA, Jesús Francisco. Real-life/real-time elderly fall detection with a triaxial accelerometer[J]. Sensors, Volume 18, Issue 4，pp.1101,2018.
[3] Wilk B, Augustyn M, Wilk G. Algorithm for human fall detection based on acceleration measurement[C]. Signal Processing: Algorithms, Architectures, Arrangements, and Applications. Poznan,pp.13-17,2020.
[4] Steven Eyobu O, Han D S. Feature representation and data augmentation for human activity classification based on wearable IMU sensor data using a deep LSTM neural network[J].Sensors, Volume 18, Issue 9,pp. 2892,2018.
[5] Ramirez H, Velastin S A, Meza I, et al. Fall detection and activity recognition using human skeleton features[J]. IEEE Access, Volume 9,pp. 33532-33542,2021.
[6] Chen E Q, Bai X, Gao L, et al. A spatiotemporal heterogeneous two-stream network for action recognition[J]. IEEE Access,Volume 7,pp.57267-57275,2019.
[7] NÚÑEZ-MARCOS, Adrián; AZKUNE, Gorka; ARGANDA-CARRERAS, Ignacio. Vision‐based fall detection with convolutional neural networks[J]. Wireless communications and mobile computing, Volume 2017, Issue 1,pp.9474806,2017.
[8] KWOLEK, Bogdan; KEPSKI, Michal. Human fall detection on embedded platform using depth maps and wireless accelerometer[J]. Computer methods and programs in biomedicine, Volume 117, Issue 3, pp.489-501,2014.
[9] ALLA, Ildi, et al. From Sound to Sight: Audio-Visual Fusion and Deep Learning for Drone Detection[C]. In: Proceedings of the 17th ACM Conference on Security and Privacy in Wireless and Mobile Networks, pp. 123-133,2024