随着社会老龄化进程的加快,行人摔倒事故成为了一个严重的社会问题。本文围绕基于改进YOLOv8的行人摔倒检测研究展开,针对原始YOLOv8在行人摔倒检测任务中存在的不足,提出了YOLOv8-RFAConv-Diou-Inner-Focaler模型。该模型通过RFA Conv...随着社会老龄化进程的加快,行人摔倒事故成为了一个严重的社会问题。本文围绕基于改进YOLOv8的行人摔倒检测研究展开,针对原始YOLOv8在行人摔倒检测任务中存在的不足,提出了YOLOv8-RFAConv-Diou-Inner-Focaler模型。该模型通过RFA Conv卷积操作,在卷积层引入注意力机制,增强对重要特征的关注,提高特征的表达能力。引入Diou损失函数解决传统损失函数在目标重叠和尺度变化时的不足以及Inner-Focaler损失函数动态调整损失权重。通过实验验证:本文提出的改进算法在行人摔倒检测任务中取得了显著的性能提升,对比原始算法(YOLOv8算法)在平均精度上提高了4.8%。With the acceleration of social aging, pedestrian falling accidents have become a serious social problem. This article focuses on the research of pedestrian fall detection based on improved YOLOv8. In response to the shortcomings of the original YOLOv8 in pedestrian fall detection tasks, a YOLOv8-RFAConv-Diou-Inner-Focaler model is proposed. This model introduces attention mechanism in the convolutional layer through RFA Conv convolution operation, enhancing the attention to important features and improving the expression ability of features. Introducing the Diou loss function to address the shortcomings of traditional loss functions in target overlap and scale changes, as well as dynamically adjusting loss weights using the Inner Focaler loss function. Through experimental verification, the improved algorithm proposed in this paper has achieved significant performance improvement in pedestrian fall detection tasks, with an average accuracy improvement of 4.8% compared to the original algorithm (YOLOv8 algorithm).展开更多
文摘随着社会老龄化进程的加快,行人摔倒事故成为了一个严重的社会问题。本文围绕基于改进YOLOv8的行人摔倒检测研究展开,针对原始YOLOv8在行人摔倒检测任务中存在的不足,提出了YOLOv8-RFAConv-Diou-Inner-Focaler模型。该模型通过RFA Conv卷积操作,在卷积层引入注意力机制,增强对重要特征的关注,提高特征的表达能力。引入Diou损失函数解决传统损失函数在目标重叠和尺度变化时的不足以及Inner-Focaler损失函数动态调整损失权重。通过实验验证:本文提出的改进算法在行人摔倒检测任务中取得了显著的性能提升,对比原始算法(YOLOv8算法)在平均精度上提高了4.8%。With the acceleration of social aging, pedestrian falling accidents have become a serious social problem. This article focuses on the research of pedestrian fall detection based on improved YOLOv8. In response to the shortcomings of the original YOLOv8 in pedestrian fall detection tasks, a YOLOv8-RFAConv-Diou-Inner-Focaler model is proposed. This model introduces attention mechanism in the convolutional layer through RFA Conv convolution operation, enhancing the attention to important features and improving the expression ability of features. Introducing the Diou loss function to address the shortcomings of traditional loss functions in target overlap and scale changes, as well as dynamically adjusting loss weights using the Inner Focaler loss function. Through experimental verification, the improved algorithm proposed in this paper has achieved significant performance improvement in pedestrian fall detection tasks, with an average accuracy improvement of 4.8% compared to the original algorithm (YOLOv8 algorithm).
