Scholarlink

<p>Balancing real-time performance and accuracy in an autonomous driving perception system for target detection under complex road conditions is a challenging task. This is due to the mobility and unpredictability of vehicle targets, which results in low detection accuracy and poor real-time performance. This paper proposes propose a light-weight vehicle detection algorithm, BFS-YOLO, to achieve efficient and accurate vehicle detection in complex road scenarios. First, this paper developed a Layer Bridge Feature Pyramid Network (LBFPN) to enhance detection accuracy through multi-scale feature fusion. Second, to address the issue of the large number of parameters of C2f Module in the original network, a lightweight Fast Module is designed to replace it, reducing the overall parameters of the network and improving operational speed by eliminating redundant convolution computations in feature fusion. To further reduce parameters, this study designs a lightweight SCHead with adaptive convolution, replacing the original detection head, improving overall network accuracy and further reducing parameters. This paper evaluated BFS-YOLO on the KITTI dataset. The proposed model contains only 1.15M parameters, achieving a 61% reduction compared to the baseline. Remarkably, the detection speed has been improved by 33 FPS compared with the previous version of the model, and the <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({mAP_{50}}\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>m</mi> <mi>A</mi> <msub> <mi>P</mi> <mn>50</mn> </msub> </mrow> </math></EquationSource> </InlineEquation> has been improved to 97.6%, representing an increase of 1.4% compared with the original network, achieving a competitive performance compared with other vehicle detection algorithms.</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

BFS-YOLO: a lightweight vehicle detection algorithm with a cross-layer feature fusion structure

  • 期刊论文

  • 发表日期：

    2025年12月23日

查看全文
  • Xuecun Yang,
  • Shushan Qiang,
  • Lintao He,
  • Jingkun Gao,
  • Gaoting Zhu,
  • Chengyi Gui,
  • Ning Ren

摘要

Balancing real-time performance and accuracy in an autonomous driving perception system for target detection under complex road conditions is a challenging task. This is due to the mobility and unpredictability of vehicle targets, which results in low detection accuracy and poor real-time performance. This paper proposes propose a light-weight vehicle detection algorithm, BFS-YOLO, to achieve efficient and accurate vehicle detection in complex road scenarios. First, this paper developed a Layer Bridge Feature Pyramid Network (LBFPN) to enhance detection accuracy through multi-scale feature fusion. Second, to address the issue of the large number of parameters of C2f Module in the original network, a lightweight Fast Module is designed to replace it, reducing the overall parameters of the network and improving operational speed by eliminating redundant convolution computations in feature fusion. To further reduce parameters, this study designs a lightweight SCHead with adaptive convolution, replacing the original detection head, improving overall network accuracy and further reducing parameters. This paper evaluated BFS-YOLO on the KITTI dataset. The proposed model contains only 1.15M parameters, achieving a 61% reduction compared to the baseline. Remarkably, the detection speed has been improved by 33 FPS compared with the previous version of the model, and the \({mAP_{50}}\) m A P 50 has been improved to 97.6%, representing an increase of 1.4% compared with the original network, achieving a competitive performance compared with other vehicle detection algorithms.