Article Details

Abstract

Fall has become one of the greatest risks to public safety, and timely detection of fall is the primary prerequisite for performing effective rescue. Object Detection based on Computer Vision is currently the mainstream method for fall detection. However, due to the complex factors in the real scenes, the existing models still face problems such as low accuracy, weak robustness, difficult deployment, and so on. The lightweight fall detection model FD-EfficientDet is proposed based on EfficientDet. Firstly, the model introduces the EMA Attention Mechanism into the backbone network to enhance the understanding and representation abilities of input data, improving detection accuracy. Secondly, the EMA-Fused-MBConv module is designed for the backbone network, to accelerate model training and prevent model degradation. Finally, the SKIPS-BiFPN structure is proposed and used to construct the Neck layer, to realize feature weighted fusion from different stages of the backbone network, thereby enhancing the robustness of the model. The experimental results show that the FD-EfficientDet model improves the mAP by 17.65% compared with the EfficientDet model, and it has good advantages in robustness, lightweight, and other aspects.

Keywords

fall detection; EfficientDet; lightweight; robustness; accuracy

References

1. [1] Wang Lan, Sun Bo, Long Changyu, etc A method for internal parameter calibration of linear array cameras based on collinear feature points [J]. Infrared and Laser Engineering, 2015, 44 (6): 1878-1883
2. [2] GUO R H， LI H，HAN D L，et al. Feasibility Analysis of Using Channel State Information (CSI) Acquired from Wi-Fi Routers for Construction Worker Fall Detection [J].International Journal of Environmental Research and Public Health，2023，20（6）：4998-4998.
3. [3] PENNONE J，AGUERO F N，MARTINI M D，et al. Fall prediction in a quiet standing balance test via machine learning： Is it possible? [J].PloS one，2024，19（4）：e0296355-e0296355.
4. [4] LI Y，HO K C，POPESCU M. A microphone array system for automatic fall detection [J].IEEE Transactions on Biomedical Engineering，2012，59（5）：1291-1301.
5. [5] VINEETH C，ANUDEEP J，KOWSHIK G，et al. An enhanced, efficient and affordable wearable elderly monitoring system with fall detection and indoor localization [J].International Journal of Medical Engineering and Informatics，2021，13（3）：254-268.
6. [6] DEDABRISHVILI M，MAMAIASHVILI N，MATIASHVILI I. Fall Detection System based on iOS Smartphone Sensors [J].Journal of Technical Science and Technologies，2024，8（1）：35-44.
7. [7] KUMAR V S，ACHARYA K G，SANDEEP B，et al. Wearable Sensor-BASed Human Fall Detection Wireless system [C]//Wireless Communication Networks and Internet of Things：Select Proceedings of ICNETS2，Volume VI.Singapore：Springer，2019：217-234.
8. [8] MEHMOOD A，NADEEM A，ASHRAF M，et al. A Novel Fall Detection Algorithm for Elderly Using SHIMMER Wearable Sensors [J].Health and Technology，2019，9（4）：631-646.
9. [9] MIN W D，CUI H，RAO H，et al. Detection of Human Falls on Furniture Using Scene Analysis Based on Deep Learning and Activity Characteristics [J].IEEE Access，2018：9324-9335.
10. [10] ZHAO Z Y，MA J W，MA C，et al. An Improved Faster R-CNN Algorithm for Pedestrian Detection [C]//2021 11th international conference on information technology in medicine and education (ITME).Wuyishan：IEEE，2021：76-80.
11. [11] XU T，CHEN J，LI Z Y. Fall Detection Based on Person Detection and Multi-Target Tracking [C]//2021 11th International Conference on Information Technology in Medicine and Education (ITME).Wuyishan：IEEE，2021：60-65.
12. [12] Xie Hui, Shi Logistics, Qi Yuxiao, etc A spatiotemporal fall detection algorithm based on attention mechanism subnetworks [J]. Computer and Modernization, 2022 (3): 70-75+81
13. [13] CHEN Y，LI W T，WANG LU，et al. Vision-Based Fall Event Detection in Complex Background Using Attention Guided Bi-Directional LSTM [J].IEEE Access，2020：161337-161348.
14. [14] REN S Q，HE K M，GIRSHICK R，et al. Faster R-CNN: Towards real-time object detection with region proposal networks [J].IEEE transactions on pattern analysis and machine intelligence，2017，39（6）：1137-1149.
15. [15] HE K M，GKIOXARI G，DOLLAR P，et al. Mask R-CNN [EB/OL].[2024-07-08].https://openaccess.thecvf.com/content_iccv_2017/html/He_Mask_R-CNN_ICCV_2017_paper.html.
16. [16] TAN M X，PANG R M，LE Q V. Efficientdet: Scalable and Efficient Object Detection [C]//2020 IEEE/CVF conference on computer vision and pattern recognition.Seattle：IEEE，2020：10778-10787.
17. [17] REDMON J，FARHADI A. Yolov3: An Incremental Improvement [J].arXiv:1804.02767 [cs.CV].[2024-07-08].https://arxiv.org/abs/1804.02767.
18. [18] OUYANG D，HE S，ZHANG G Z，et al. Efficient Multi-Scale Attention Module with Cross-Spatial Learning [C]//ICASSP 2023-2023 IEEE International Conference on Acoustics，Speech and Signal Processing (ICASSP).Rhodes Island：IEEE，2023：1-5.

About the authors

Mou Xuenan (1998-), male, Han ethnicity, from Bengbu, Anhui, holds a master's degree and specializes in object detection; Yu Jieyi (1997-), female, Han ethnicity, from Zhongshan, Guangdong, holds a master's degree and specializes in classification prediction; Wang Yin (1982-), male, Han ethnicity, from Zhoukou, Henan Province, lecturer, PhD, research direction: digital healthcare; Hu Wenjun (1977-), male, Han ethnicity, from Xuancheng, Anhui Province, professor, PhD, research direction: machine learning.