摘要
在人工智能领域中,车辆自动驾驶是一项重要研究内容,交通标志检测是其中重要一环。为了实现对交通标志的精确检测,基于YOLOv3提出一种改进的交通标志检测算法。首先,提出长期分割傅里叶变化模型(LSFM)完成5帧图片关联分割问题,在YOLOV3的DBL层后接入LSFM层,对输入的特征进行时序上的关联分析,并使用图像补丁方法,提升对交通标志特征提取能力,提高目标召回率。其次,提出一种基于傅里叶变换的分割选择算法(SFM),主要是通过将图片编码,对图像进行下采样,将图片矩阵变换成一维矩阵,作为一组信号特征,通过傅里叶变换,分析不同的频谱特性,再通过阈值完成对不同区域的分割划分,大大降低目标误报率,提升交通标志检测精度。实验结果表明,改进后的算法AUC最高提升15%,可以更高效地完成对交通目标检测。
In the field of artificial intelligence,vehicle automatic driving is an important research content,and traffic sign detection is an important part of it.In order to realize the accurate detection of traffic signs,this paper proposes an improved traffic sign detection algorithm based on YOLOv3.Firstly,this article proposes the Long Term Segmentation Fourier Transform Model(LSFM)to solve the problem of 5-frame image correlation segmentation.After the DBL layer of YOLOV3,the LSFM layer is connected to perform temporal correlation analysis on the input features.The image patch method is used to improve the ability to extract traffic sign features and improve the target recall rate.Secondly,this article proposes a segmentation selection algorithm based on Fourier transform(SFM),which mainly encodes the image,downsampling the image,transforming the image matrix into a one-dimensional matrix as a set of signal features.Through Fourier transform,different spectral characteristics are analyzed,and then the segmentation of different regions is completed through threshold,greatly reducing the target false alarm rate and improving the accuracy of traffic sign detection.The experimental structure shows that the AUC of the improved algorithm can be improved up to 15%.The traffic target detection can be completed more efficiently.
作者
江磊
叶旺
王建华
JIANG Lei;YE Wang;WANG Jian-hua(School of Mechanical and Automobile Engineering,Shanghai University of Engineering Science,Shanghai 201620,China;Hangzhou State Control Electric Power Technology Co.,LTD,Hangzhou Zhejiang 310000,China)
出处
《计算机仿真》
2025年第2期140-146,共7页
Computer Simulation
关键词
交通标志
检测算法
仿真
Traffic sign
Detection algorithm
Simulation
