基于图像聚类的交通标志CNN快速识别算法

Title:

CNN-based image clustering algorithm for fast recognition of traffic signs

作者:

伍锡如^1,2, 雪刚刚^1,2

1. 桂林电子科技大学 电子工程与自动化学院, 广西 桂林 541004;
2. 桂林电子科技大学 广西自动检测重点实验室, 广西 桂林 541004

Author(s):

WU Xiru^1,2, XUE Ganggang^1,2

1. College of Electronic Engineering and Automation, Guilin University of Electronic Technology, Guilin 541004, China;
2. Guangxi Key Laboratory of Automatic Detection, Guilin University of Electronic Technology, Guilin 541004, China

关键词:

交通标志; 模式识别; 图像预处理; 图像聚类; 样本优化; 深度学习; 卷积神经网络; 智能汽车

Keywords:

traffic sign; pattern recognition; image preprocessing; image clustering; sample optimization; deep learning; convolutional neural network; intelligent vehicle

分类号:

TP391.4

DOI:

10.11992/tis.201806026

摘要:

为了提高交通标志图像识别的准确性和实时性，提出一种基于图像聚类的交通标志CNN快速识别算法。利用图像聚类算法对原始数据集进行样本优化；采用多种图像预处理操作使样本整体质量进一步提升；构造了深度为9的CNN结构，通过多次训练得到最终的网络模型，将待识别的图像输入到CNN模型来实现自动识别。在德国交通标志数据集（German traffic sign recognition benchmark， GTSRB）和比利时交通标志数据集（Belgium traffic sign dataset， BTSD）上证明了算法的有效性，单张图片的识别速度只需0.2 s，识别精度高达98.5%以上。本算法具有识别速度快、准确率高的特点，可为智能驾驶的可靠性和安全性提供理论依据和技术支持。

Abstract:

To improve the accuracy and real-time performance of traffic signs recognition, a convolutional neural network (CNN)-based image clustering algorithm is proposed for fast recognition of traffic signs. First, The image clustering algorithm was used to optimize the original dataset; second, the overall quality of the dataset was further improved through various image preprocessing operations; finally, a nine-layer CNN structure was constructed. The CNN model was eventually obtained after several sample trainings, and an image was inputted into the CNN model for automatic recognition. The validity of the algorithm was proved based on the German traffic sign recognition benchmark (GTSRB) and Belgium traffic sign dataset (BTSD). The recognition time of a single picture was only 0.2 s, and the recognition accuracy was higher than 98.5%. The results confirm fast recognition and high accuracy rate of the proposed algorithm. It provides theoretical basis and technical support for the reliability and security of intelligent vehicle driving.