首页|基于几何透视图像预处理和CNN的全景图像交通标志识别算法

基于几何透视图像预处理和CNN的全景图像交通标志识别算法

扫码查看
为解决深度学习方法在高清全景图像中检测交通标志遇到图形处理器资源不足、小目标容易漏检、检测速度过慢等问题,采用小目标过采样训练数据生成方法、图像分块和几何透视检测预处理方法以及改进的轻量神经网络Improved-Tiny-YOLOv3,提出了 一种基于深度学习的轻量级全景图像中交通标志检测方法.并在Tsinghua-Tencent 100K数据集上进行了实验,mAP值达到92.7%,在Nvidia 1080Ti显卡上检测速度可达到20 FPS,实验结果验证了所提方法的有效性.
A PANORAMIC IMAGE TRAFFIC SIGN RECOGNITION ALGORITHM BASED ON GEOMETRIC PERSPECTIVE IMAGE PREPROCESSING AND CNN
In order to solve the problems of insufficient graphics processor resources,small targets being easily missed,and slow detection speed when the deep learning method is used to detect traffic signs in high-definition panoramic images,this paper proposes a lightweight panoramic image traffic sign detection method based on deep learning.It used a small target oversampling training data generation method,image segmentation and geometric perspective detection preprocessing method,and an improved lightweight neural network Improved-Tiny-YOLOv3.Experiments were performed on the Tsinghua-Tencent 100K data set,and the mAP value reached 92.7%,and the detection speed on the Nvidia 1080Ti graphics card reached 20FPS.The experimental results show the effectiveness of the proposed method.

Traffic sign detection and recognitionImproved-Tiny-YOLOv3Geometric perspectiveRandom cropCIoUPanoramic image

曹峻凡、张向利、闫坤、张红梅

展开 >

桂林电子科技大学广西无线宽带通信与信号处理重点实验室 广西桂林 541004

交通标志检测识别 Improved-Tiny-YOLOv3 几何透视法 随机裁剪 CIoU 全景图像

广西无线宽带通信与信号处理重点实验室2020年主任基金项目广西云计算与大数据协同创新中心项目

GXKL06200104YD1904

2024

计算机应用与软件
上海市计算技术研究所 上海计算机软件技术开发中心

计算机应用与软件

CSTPCD北大核心
影响因子:0.615
ISSN:1000-386X
年,卷(期):2024.41(7)