YOLOv5-FRNet: A Real-time Drowning Detection Method with Multi-Model Cascade

DOI: https://doi.org/10.62517/jbdc.202501210

Author(s)

Yitong Zhou1, Jingjing Wang1, Lu Li1, Wanwan Wang2

Affiliation(s)

1School of Artificial Intelligence and Big Data, Henan University of Technology, Henan, Zhengzhou, China 2IFLYTEK CO.LTD, Anhui, Hefei, China

Abstract

Anti-drowning detection is of great significance to ensuring the safety of public waters, but the existing vision-based detection methods have problems such as high target miss detection rate and insufficient extraction of behavioral features in complex scenarios, and there is a lack of public anti-drowning detection datasets. To this end, this experiment collected and disclosed 8518 drowning prevention detection dataset and use LabelImg for image annotation. Then, a real-time drowning detection method with multi-model cascade was designed and implemented. First, the effects of Faster-RCNN and YOLOv5 methods in drowning detection were compared, and the results were cascaded to form YOLOv5-FRNet. The experimental results show that in terms of recall, YOLOv5-FRNet is 0.603, which is 0.01 and 0.113 higher than Faster-RCNN and YOLOv5, respectively. In terms of comprehensive detection performance, the IoU=0.5 index: YOLOv5-FRNet (0.774) is optimized with 1.3% and 27.7% compared with Faster-RCNN (0.764) and YOLOv5 (0.606) respectively; the IoU=0.5:0.95 index: YOLOv5-FRNet model (0.603) is 25.6% compared with Faster-RCNN (0.480), significantly better than YOLOv5 (0.388). The methods proposed in this article can be applied to swimming pools and other places, providing a reference for high-precision anti-drowning detection.

Keywords

Target Detection; Anti-Drowning Detection; YOLOv5; Faster-RCNN

References

[1] Maad S, Frdoos A, Ashwaq A, et al. Deep Learning and Vision-Based Early Drowning Detection. Information, 2023, 14 (1): 52-52. [2] Yu Zhongyang, Wang Yajing. A Comparative Review of Drowning Monitoring Systems. Computer Knowledge and Technology, 2023, 19(13): 119-122+126. [3] Liu T, He X, He L, et al. A video drowning detection device based on underwater computer vision. IET Image Processing, 2023, 17(6): 1905-1918. [4] Wang Yun. Research on a Deep Learning-Based Early Warning and Monitoring Management System for Drowning Incidents. China New Technologies and Products, 2024, (01): 146-148. [5] Jing Mingtao, Yu Teng, Feng Mengyao, Yang Guowei. Research on Drowning Detection in Swimming Pools Based on Improved Mask R-CNN. Journal of Qingdao University (Engineering & Technology Edition), 2021, 36(01): 1-7+21. [6] Peng Ting, Shen Jinghu, Qiao Yu. Design of a Drowning Behavior Detection System in Swimming Pools Based on Improved Mask R-CNN. Transducer and Microsystem Technologies, 2021, 40(01): 94-97. [7] Yu Zhongyang, Yang Wenhui. A Swimming Pool Drowning Detection Algorithm Based on Human Pose Estimation. Technology Innovation and Application, 2023, 13(23): 66-70+74. [8] Chen Juan, Ge Bi, Chen Dongsheng. Research on a Reservoir Drowning Monitoring and Early Warning System. Computer Knowledge and Technology, 2024, 20(10): 5-7+14. [9] Xie Jiangna, Liu Liqiu, Zhao Jialin, Wang Yiluan. Design of a Portable Swimming Pool Drowning Alarm and Physiological Health Assessment System. Electronic Engineering & Product World, 2023, 30(03): 19-23. [10] Liu Xiangju, Shuai Tao, Jiang Shexiang. Drowning Personnel Detection Based on Improved YOLOv5. Journal of Shaanxi University of Technology (Natural Science Edition), 2024, 40(03): 35-43. [11] Liu Songtao, Lü Hui, Liu Xiangling. Analysis of the Diagnostic Value of Artificial Intelligence Technology in Retinal Vein Occlusion. Recent Advances in Ophthalmology, 2025, 45(01): 46-49. DOI: 10.13389/ j. cnki. rao. 2025. 0009.