Scholarlink

<p>Driver behavior detection is crucial to improving traffic safety and protecting driver health. Traditional driver behavior monitoring methods often rely on complex models with high computing resources, which are difficult to meet the requirements of real-time and lightweight. This study proposes a full-layer aggregation network (FAN) based on the YOLO (You Only Look Once) architecture, and constructs a lightweight driver behavior detection algorithm called FAN-YOLO. FAN-YOLO combines the YOLO object detection framework and lightweight network structure. Based on the standard YOLO framework, it changes the network structure of the neck layer path aggregation network (PAN) and adopts FAN to alleviate the problems of complex PAN path interactions and information redundancy. At the same time, a new feature fusion module is proposed in the backbone network, which we call RC2A, to extract better feature information. Then, the spatial pyramid pooling (SPP) network is optimized to obtain SPPF_WD, so that FAN-YOLO can detect driver behavior more accurately. FAN-YOLO is experimented on two datasets of abnormal driver behavior. The experimental results show that the number of parameters of FAN-YOLO is only 2.2M, which is 18.5% lower than the SOTA model YOLOv8-N in YOLO. The calculation amount is 6.2G, which is 10% lower. On the improved datasets State Farm v2 and YawDD v2, <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11554_2025_1720_Article_IEq1.gif" Format="GIF" Height="21" Rendition="HTML" Resolution="72" Type="Linedraw" Width="45" /> </InlineMediaObject> <EquationSource Format="TEX">\(AP_{50}^{val}\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>A</mi> <msubsup> <mi>P</mi> <mrow> <mn>50</mn> </mrow> <mrow> <mi mathvariant="italic">val</mi> </mrow> </msubsup> </mrow> </math></EquationSource> </InlineEquation> can reach 86.8% and 97.5%, 1.3% higher than YOLOv8-N. The response speed is 1.9ms, which is 29.6% higher.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

FAN-YOLO: Real-time driver behavior detection based on full-layer aggregation network of YOLO

  • 期刊论文

  • 发表日期：

    2025年7月1日

查看全文
  • Xinsu Gan,
  • Lidong Huang,
  • Yan Yuan,
  • Yan Deng

摘要

Driver behavior detection is crucial to improving traffic safety and protecting driver health. Traditional driver behavior monitoring methods often rely on complex models with high computing resources, which are difficult to meet the requirements of real-time and lightweight. This study proposes a full-layer aggregation network (FAN) based on the YOLO (You Only Look Once) architecture, and constructs a lightweight driver behavior detection algorithm called FAN-YOLO. FAN-YOLO combines the YOLO object detection framework and lightweight network structure. Based on the standard YOLO framework, it changes the network structure of the neck layer path aggregation network (PAN) and adopts FAN to alleviate the problems of complex PAN path interactions and information redundancy. At the same time, a new feature fusion module is proposed in the backbone network, which we call RC2A, to extract better feature information. Then, the spatial pyramid pooling (SPP) network is optimized to obtain SPPF_WD, so that FAN-YOLO can detect driver behavior more accurately. FAN-YOLO is experimented on two datasets of abnormal driver behavior. The experimental results show that the number of parameters of FAN-YOLO is only 2.2M, which is 18.5% lower than the SOTA model YOLOv8-N in YOLO. The calculation amount is 6.2G, which is 10% lower. On the improved datasets State Farm v2 and YawDD v2, \(AP_{50}^{val}\) A P 50 val can reach 86.8% and 97.5%, 1.3% higher than YOLOv8-N. The response speed is 1.9ms, which is 29.6% higher.