摘 要：摔倒是影响老人生命安全的重要问题之一，为提高检测准确率，将 YOLOv5 模型应用于摔倒检测并做了改进。首先，用 K-means 聚类算法得到更符合目标形态的 anchor 长宽比例，使边界框更精确，提高摔倒检测准确率。其次，用 EIoU 损失函数替换 CIoU 损失函数，加快收敛速度，使目标定位更准确。实验结果表明，改进后的 YOLOv5 模型检测效果较好，准确率达到99.1%，mAP 值达到 99.3%，能够更好地满足摔倒检测的要求。
关键词：改进 YOLOv5 算法；人体摔倒检测；K-means 聚类；EIoU 损失函数
DOI:10.19850/j.cnki.2096-4706.2023.04.031
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2023）04-0121-04
Improved YOLOv5 Human Fall Detection
ZHANG Jing
(College of Computer Science and Technology, Taiyuan Normal University, Jinzhong 030619, china)
Abstract: Falling is one of the important problems affecting the life safety of the elderly. In order to improve the detection accuracy, YOLOv5 model is applied to fall detection and improved. Firstly, K-means clustering algorithm is used to get the ratio of the length and width of the anchor which is more in line with the target shape, and it makes the bounding box more accurate and improves the accuracy of fall detection. Secondly, it replaces the CIoU loss function with EIoU loss function to speed up the convergence and make the target location more accurate. The experimental results show that the improved YOLOv5 model has a better detection effect, with an accuracy rate of 99.1% and the mAP value reaches 99.3%, which can better meet the requirements of fall detection.
Keywords: improved YOLOv5 algorithm; human fall detection; K-means clustering; EIoU loss
参考文献：
[1] 国家统计局 . 第七次全国人口普查主要数据情况 [EB/OL].（2021-05-11）.http://www.stats.gov.cn/ztjc/zdtjgz/zgrkpc/dqcrkpc/ggl/202105/t20210519_1817693.html.
[2] 中国老年人跌倒风险评估专家共识（草案） [J]. 中国老年保健医学，2019，17（04）：47-48+50.
[3] 崔建英，刘燕，熊英 . 老年综合评估护理方案在老年人院外跌倒防范中的应用 [J]. 护理研究，2020，34（24）：4473-4475.
[4] ALESIN A，OSANLOU A，MAW S W. A low budget multifunctional wearable device for motion and falls detection [C]//2018IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering （EIConRus）.IEEE，2018：1850-1855.
[5] LIN C L，CHIU W C，CHEN F H，et al. Fall Monitoring for the Elderly Using Wearable Inertial Measurement Sensors on Eyeglasses [J].IEEE Sensors Letters，2020，4（6）：1-4.
[6] YANG X，XIONG F，SHAO Y，et al. WmFall：WiFi-based multistage fall detection with channel state information [J].International Journal of Distributed Sensor Networks，2018，14（10）：1-10.
[7] WANG X，JIA K. Human Fall Detection Algorithm Based on YOLOv3 [C]//2020 IEEE 5th International Conference on Image， Vision and Computing（ICIVC）.IEEE，2020：50-54.
[8] 周洪成，杨娟，徐志国 . 基于 YOLOv5 算法的人体跌倒检测系统设计 [J]. 金陵科技学院学报，2022，38（2）：22-29.
[9] 马露，裴伟，朱永英，等.基于深度学习的跌倒行为识别 [J]. 计算机科学，2019，46（9）：106-112.
[10] ZHENG Z，WANG P，REN D，et al. Enhancing Geometric Factors in Model Learning and Inference for Object Detection and Instance Segmentation [J].IEEE Transactions on Cybernetics，2022，52（8）：8574-8586.
[11] ZHANG Y F，REN W，ZHANG Z，et al. Focal and Efficient IOU Loss for Accurate Bounding Box Regression [J].arXiv： 2101.08158 [cs.CV].（2021-01-20）.https://arxiv.org/abs/2101.08158v2.
作者简介：张靖（1997—），女，汉族，陕西宝鸡人，硕士研究生在读，研究方向：智能数据分析、目标检测。