| 面向老年人跌倒检测的改进YOLOv7算法 |
 点此下载全文 |
| 引用本文：易峥荣,潘昊,谢冰倩.面向老年人跌倒检测的改进YOLOv7算法[J].计算技术与自动化,2026,(1):9-17 |
| 摘要点击次数: 299 |
| 全文下载次数: 274 |
|
|
| 中文摘要:随着人口老龄化的问题日益凸显，室内跌倒事件的发生率也随之上升，老人跌倒成为家庭和社会关注的重点问题。针对目前的跌倒检测模型往往需要大量的计算资源和参数配置，且在复杂环境中这些跌倒检测算法的准确率仍然较低等问题，提出了一种改进型的YOLO-GDDH的跌倒检测算法。首先，借鉴YOLOv9提出的广义高效层聚合网络（GELAN），在原始骨干网络中加入RepNCSPELAN4模块，保证网络轻量级的基础上提升网络推理速度和准确度。其次，引用一种先进的聚集和分布（GD）机制来改进原有的路径聚合网络（PANet），增强颈部网络的信息融合能力。最后，采用动态检测头进行检测结果输出，加快网络收敛速度。实验结果表明，改进算法在测试集上的均值平均精度 mAP@ 0.5 达到 97.8%，优于基线 YOLOv7-tiny；同时，模型仅有6.4 M的参数量和14.3 B的计算量。相较于其他主流的轻量化目标检测模型，YOLO-GDDH 在保证模型轻量化的同时具有较高的检测精度，验证了本文所提方法的有效性。 |
| 中文关键词:跌倒检测；YOLOv7-tiny；轻量化网络；聚集和分布机制 动态检测头 |
| |
| An Improved YOLOv7 Algorithm For Fall Detection in Elderly People |
|
|
| Abstract:Elderly falls have become a major issue for families and society as a result of the aging population’s increasing prevalence and the rise in indoor fall incidents. This paper proposes an improved fall detection algorithm for YOLO-GDDH, aiming at the issues that current fall detection models often require a large number of computational resources and parameter configurations, the accuracy of these fall detection algorithms remains relatively low in complex environments. First, the RepNCSPELAN4 module is added to the original backbone network, building on the generalized efficient layer aggregation network (GELAN) developed by YOLOv9, to enhance the network inference speed and accuracy while ensuring a lightweight network. Second, the original path aggregation network (PANet) is modified using a better gather-and-distribute mechanism (GD) mechanism. This improves the neck network's information fusion capability. Finally, the dynamic head is used to output detection results to accelerate the network convergence speed. The experimental results show that the improvement algorithm achieves 97.8% mean average precision mAP@ 0.5 on the test set, which is better than the baseline YOLOv7-tiny. meanwhile, the model has only 6.4 M parametric quantities and 14.3 B computational volume. in contrast to other popular lightweight target detection models, YOLO-GDDH has a high detection precision while maintaining the lightweight model, confirming the efficacy of the technique presented in this paper. |
| keywords:fall detection YOLOv7-tiny lightweight network gather-and-distribute mechanism DyHead |
| 查看全文 查看/发表评论 下载pdf阅读器 |
|
|
|
位访问者,今日访问
