摘要
交通标志识别是智能驾驶的关键技术,要满足识别准确率高和识别速度快的要求。为了提升交通标志的识别准确率和识别速度,提出基于卷积神经网络的交通标志识别算法,设计了一种准确率高、速度快的识别模型用于交通标志识别。该模型使用了改进的Inception模块以及多尺度特征融合方式增强网络的特征提取能力,采用批量归一化来加速网络的训练,采用全局平均池化减小模型的参数量。在GTSRB数据集上进行训练测试,识别模型的准确率达到99.6%,识别每张图片的时间为0.22ms,实验结果表明识别模型的识别准确率高,识别速度快。通过自对比实验,验证了识别模型的结构优势。与其他交通标志识别方法在GTSRB数据集上进行对比实验,识别模型的识别性能优于其他识别方法。
Traffic sign recognition is a key technology for intelligent driving,which should meet the requirements of high recognition accuracy and fast recognition speed.In order to improve the recognition accuracy and recognition speed of traffic signs,a traffic sign recognition algorithm based on convolutional neural network is proposed,and a recognition model with high accuracy and fast speed is designed for traffic sign recognition.The model uses improved Inception module and multi-scale feature fusion to enhance the fea-ture extraction ability of the network,batch normalization to accelerate the training of the network,and global average pooling to re-duce the number of parameters of the model.Training and test are conducted on the GTSRB dataset,and the recognition model achieves an accuracy of 99.6%,and the time to recognize each image is 0.22ms.The experimental results show that the recognition model has high recognition accuracy and fast recognition speed.The structural advantages of the recognition model are verified through self-comparison experiments.Comparison experiments with other traffic sign recognition methods on the GTSRB dataset show that the recognition performance of the recognition model is better than other recognition methods.
作者
张小雪
黄巍
ZHANG Xiao-xue;HUANG Wei(School of Computer Science and Engineering,Wuhan Institute of Technology,Wuhan 430000,Hubei)
出处
《电脑与电信》
2022年第7期1-5,9,共6页
Computer & Telecommunication
基金
中国高校产学研创新基金,2020ITA05020。
