School of Mechanical and Electronic Engineering,East China University of Technology,Nanchang 330013,China
Show less
文章历史+
Received
Published
Revised
2025-01-13
2026-02-08
2025-10-22
Online Date
2026-09-21
PDF (8134K)
摘要
针对当前基于视觉的跌倒检测方法无法较好平衡检测精度与响应速度、所设实验环境单一的问题,提出了一种基于YOLOv11n改进的轻量级跌倒检测算法MWA-YOLO。将轻量级的通道注意力机制MLCA(mixed local channel attention)引入C3K2模块,增强通道特征的表现力以减少通道分割导致的信息丢失。引入WIoU替换原模型中的CIoU作为新的边界框损失函数,提升模型在复杂环境中边界框的定位精度。将网络结构中的部分标准卷积Conv层替换为改进后的下采样ADown模块,增强模型对有效信息捕捉能力的同时保持轻量化。实验结果表明,MWA-YOLO跌倒检测算法较YOLOv11n在mAP@0.5指标上从92.3%提升到了93.5%,参数量和计算量仅有2.2×106和6.0×109,可部署在计算资源有限的硬件设备中,用于日常生活场景中的跌倒检测。
Abstract
To address the challenges of balancing detection accuracy and response speed in vision-based fall detection methods,as well as the limitations posed by a single experimental environment,a lightweight fall detection algorithm based on an improved YOLOv11n,termed MWA-YOLO,was proposed. The algorithm incorporated a lightweight channel attention mechanism mixed local channel attention(MLCA) into the C3K2 module to enhance channel feature representation and mitigate information loss due to channel splitting. Additionally,the WIoU loss function was introduced to replace the original CIoU,improving bounding box localization accuracy in complex environments. Several standard convolutional layers in the network were substituted with the advanced downsampling ADown module to bolster the model’s ability to capture relevant features while maintaining its lightweight characteristics. Experimental results demonstrate that MWA-YOLO achieves an improvement in the mAP@0.5 metric from 92.3% to 93.5% when compared to YOLOv11n,with a model size of only 2.2×106 parameters and 6.0×109 FLOPs. This makes the proposed method suitable for deployment on hardware with limited computational resources,making it ideal for fall detection in real-life scenarios.
YIN Jiao, LUO Xian-xi, LIU Shu-bo.
Improved Fall Detection Algorithm Based on YOLOv11n[J].
Science Technology and Engineering, 2026, 26(4): 1586-1594 DOI:10.12404/j.issn.1671-1815.2500294
在YOLO系列算法中,数据集是决定模型性能的关键因素之一。在训练过程中,数据集为模型提供足够的检测特征和背景信息,帮助其更好的学习目标特征,同时包含一定程度的干扰信息帮助模型提高鲁棒性。为使得训练后的模型在不同场景中都能拥有较好的准确性和泛化能力,因此数据集需要具备多样化特性。选取3个公开数据集:FDD(fall detection dataest)[15]、MCF(multiple cameras fall dataset)[16]和URFD(UR fall detection dataset)[17]。同时为保证数据集中场景的多样性,使模型能从中学习到更多日常生活场景中发生跌倒事件的特征,从网络平台收集许多室内与室外的跌倒与非跌倒图像,将这些图像与上述3个公开数据集相混合,使用LabelImg软件对3个数据集进行手工标注,设定3种类别:p_fallen(跌倒)、p_stand(站立/弯腰)、p_sit(坐/蹲坐)。标记完成的图像划分为训练集、验证集、测试集3种,其中训练集3 486幅。数据集中包含不同身高、体型、衣服穿着的人,不同角度的拍摄,以及不同的场景和光照设计,保证了数据的多样性。
AlamE, SufianA, DuttaP, et al. Vision-based human fall detection systems using deep learning: a review[J]. Computers in Biology and Medicine, 2022, 146: 105626.
WangLi, ZhaoYanyang, LiuXuefeng, et al. The research progress of the fall detection technology based on the inertial sensor[J]. Science Technology and Engineering, 2017, 17(32): 205-213.
XieZongyuan, MaHongyan, LiShengyan, et al. Elderly fall detection based on lightweight YOLOv5 and dual cameras[J]. Science Technology and Engineering, 2024, 24(33): 14330-14338.
FuNana, LiuDaming, ChengXiaoting, et al. Fall detection algorithm based on lightweight OpenPose model[J]. Transducer and Microsystem Technologies, 2021, 40(11): 131-134, 138.
XiongMingfu, LiJiahui, XiongJiefan, et al. Pedestrian fall detection method based on optical flow reconstruction and depth pose feature[J]. Science Technology and Engineering, 2022, 22(35): 15688-15696.
[12]
HwangH, KimD, KimH. FD-YOLO: a YOLO network optimized for fall detection[J]. Applied Sciences, 2025, 15(1): 453.
[13]
KhanH, UllahI, ShabazM, et al. Visionary vigilance: optimized YOLOV8 for fallen person detection with large-scale benchmark dataset[J]. Image and Vision Computing, 2024, 149: 105195.
ZhuZhouhua, HouZhijie, TianChengyuan, et al. Distracting driving detection and identification based on an improved YOLOv8-pose[J]. Electronic Measurement Technology, 2024, 47(15): 135-143.
[18]
WanD, LuR, ShenS, et al. Mixed local channel attention for object detection[J]. Engineering Applications of Artificial Intelligence, 2023, 123: 106442.
[19]
WangK, ZhouH, WuH, et al. RN-YOLO: a small target detection model for aerial remote-sensing images[J]. Electronics, 2024, 13(12): 2383.
[20]
WangC Y, YehI H, Mark LiaoH Y. YOLOv9: learning what you want to learn using programmable gradient information[C]// European Conference on Computer Vision. Berlin: Springer, 2025: 1-21.
[21]
AdhikariK, BouchachiaH, Nait-CharifH.Activity recognition for indoor fall detection using convolutional neural network[C]//2017 Fifteenth IAPR International Conference on Machine Vision Applications (MVA). New York: IEEE, 2017: 81-84.
[22]
AuvinetE, RougierC, MeunierJ, et al. Multiple cameras fall dataset[J]. Health Care Management Science, 2010, 1350: 24.
[23]
KwolekB, KepskiM. Human fall detection on embedded platform using depth maps and wireless accelerometer[J]. Computer Methods and Programs in Biomedicine, 2014, 117(3): 489-501.
[24]
OuyangD, HeS, ZhangG, et al. Efficient multi-scale attention module with cross-spatial learning[C]// ICASSP 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). New York: IEEE, 2023: 1-5.
[25]
HuJ, ShenL, SunG. Squeeze-and-excitation networks[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2018: 7132-7141.
[26]
WooS, ParkJ, LeeJ Y, et al. CBAM: convolutional block attention module[C]// Proceedings of the European Conference on Computer Vision (ECCV). Berlin: Springer, 2018: 3-19.
[27]
LiuW, AnguelovD, ErhanD, et al. SSD: single shot multibox detector[C]// Computer Vision-ECCV 2016: 14th European Conference, Amsterdam: Springer, 2016: 21-37.