Journal of Computer Applications

    Next Articles

Multi-target detection algorithm for traffic intersection images based on YOLOv9

  

  • Received:2024-07-19 Revised:2024-11-04 Online:2024-11-19 Published:2024-11-19

基于YOLOv9的交通路口图像的多目标检测算法

廖炎华1,鄢元霞2,潘文林3   

  1. 1. 云南民族大学
    2. 国网四川省电力公司成都市新津供电分公司
    3. 云南民族大学数学与计算机科学学院
  • 通讯作者: 廖炎华
  • 基金资助:
    云南彝绣数字化关键技术研究

Abstract: Aiming at the problem of the complex traffic intersection images, the difficulty in detecting small targets, and the tendency for occlusion between targets, as well as the color distortion, noise, and blurring caused by changes in weather and lighting, a multi-target detection algorithm ITD-YOLOv9 for traffic intersection images based on YOLOv9 (You Only Look Once v9) was proposed. Firstly, the CoT-CAFRNet (Chain-of-Thought Prompted Content-Aware Feature Reassembly Network) image enhancement network was designed to improve image quality and optimize input features. secondly, the Channel Adaptive Feature Fusion (iCAFF) module was developed to enhance feature extraction for small targets and overlapped and occluded targets. thirdly, the BiHS-FPN (Bi-directional High-level Screening Feature Pyramid Network) structure was proposed to enhance multi-scale feature fusion capability. Finally, the IF-MPDIoU (Inner-Focaler-Minimum Point Distance based Intersection over Union) loss function was designed to enhance generalization ability by adjusting variable factors, focusing on key samples. Experimental results show that, on the self-made dataset and SODA10M dataset, ITD-YOLOv9 can achieves 83.3% and 56.3% detection accuracy and 64.8frame/s and 57.4frame/s detection speeds, respectively. Compared with YOLOv9 algorithm, the detection accuracy is improved by 3.9 and 2.7 percentage points, effectively achieving multi-target detection at traffic intersections.

Key words: YOLOv9, traffic intersection detection, CoT-CAFRNet, adaptive fusion, BiHS-FPN, IF-MPDIoU loss function

摘要: 针对交通路口图像复杂,小目标难测且目标之间易遮挡,天气和光照变化又引发颜色失真、噪声多和模糊等问题,提出一种基于YOLOv9(You Only Look Once v9)的交通路口图像的多目标检测算法ITD-YOLOv9。首先,设计CoT-CAFRNet(Chain-of-Thought Prompted Content-Aware Feature Reassembly Network)图像增强网络,提升图像质量,优化输入特征。其次,加入通道自适应特征融合(iCAFF)模块,增强小目标及重叠遮挡目标的提取能力。然后,提出特征融合金字塔结构BiHS-FPN(Bi-directional High-level Screening Feature Pyramid Network),增强多尺度特征融合能力。最后,设计IF-MPDIoU(Inner-Focaler-Minimum Point Distance based Intersection over Union)损失函数,通过调整变量因子,聚焦关键样本,增强泛化能力。实验结果表明:在自制数据集和SODA10M数据集上,ITD-YOLOv9的检测精度分别为83.8%和56.3%,检测速度分别为64.8frame/s和57.4frame/s。与YOLOv9算法相比,检测精度分别提升3.9和2.7个百分点,有效实现交通路口下的多目标检测。

关键词: YOLOv9, 交通路口检测, CoT-CAFRNet, 自适应融合, BiHS-FPN, IF-MPDIoU损失函数

CLC Number: