摘要
受到跌倒时动作变化快、姿态多和复杂环境的影响,跌倒检测算法会出现误检、漏检和检测速度慢的问题。为解决上述问题,提出一种基于改进的YOLOv5算法,进行居家环境下的实时跌倒检测。使用RepVGG对YOLOv5的Backbone进行优化,增强骨干网络特征提取能力;在网络中添加卷积注意力机制,强化模型对重要特征的关注;在特征融合部分引入加权双向特征金字塔网络并简化,充分融合不同尺度特征。实验结果表明,与未改进前相比,该方法在精度、召回率和平均精度上分别提高3.43%、1.41%和3.1%,优化了检测效果,更好地满足实际使用要求。
Due to the rapid change of human movement,many posture and complex environment,the fall detection algorithm will have the problems of false detection,missing detection and slow detection speed.To solve the above problems,an improved YOLOv5 algorithm is proposed for real⁃time fall detection in home environment.RepVGG is used to optimize the Backbone of YOLOv5 to enhance the feature extraction capability of the backbone network.The convolutional attention mechanism is added to the network to strengthen the model’s attention to important features.The weighted bidirectional feature pyramid network is introduced into the feature fusion part and simplified to integrate features of different scales more fully.The experimental results show that the accuracy,recall rate and average accuracy of the method are improved by 3.43%,1.41%and 3.1%respectively,which optimizes the detection effect and better meets the practical requirements.
作者
冯梓文
冯云霞
FENG Ziwen;FENG Yunxia(School of Information Science and Technology,Qingdao University of Science and Technology,Qingdao 266061,China)
出处
《电子设计工程》
2025年第1期1-6,共6页
Electronic Design Engineering
基金
国家自然科学基金资助项目(61702135,61806107)
山东省自然科学基金面上项目(ZR2023MF088)。
