| 摘要: |
| 针对特定场景交通标志精度低与识别速度慢的问题，基于交通标志边缘信息与卷积神经网络，提出了一种交通标志图像识别T-YOLO算法。该算法基于YOLOv2算法检测思想，融合残差网络、卷积层填充0等结构，下采样舍弃池化层改用卷积层，并提取边缘信息与上采样以提升精度，设计7层特征提取网络以缩短识别速度，随后使用Softmax函数归一化实现多分类，并采用批量归一化、多尺度训练等方法缩短训练时间。实验表明，该算法真实有效，图形处理单元（Graphic Processing Unit，GPU）平台上最快检测速度13.69 ms/frame，每帧缩短9.51 ms，最高平均准确率97.3%，提高7.1%，满足实时高精度识别要求。与其他算法相比，该算法在交通标志识别速度与精度方面均有大幅提高，更加适用于现实场景，更贴近车载嵌入式系统。 |
| 关键词: 智能交通系统 车载嵌入式系统 交通标志识别 深度学习 T-YOLO 多尺度训练 残差网络 |
| DOI： |
|
| 基金项目:国家自然科学基金资助项目（61671095,61702085,61701065,61771067）；重庆市研究生教育教学改革研究重点项目（yjg192019） |
|
| A Novel Traffic Sign Recognition Algorithm Based on Deep Learning |
| CHEN Changchuan,WANG Haining,ZHAO Yue,WANG Yanping,LI Lianjie,LI Kui,ZHANG Tianqi |
| (1.School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;2.School of Information Science and Engineering,Shandong University,Qingdao 266237,China) |
| Abstract: |
| For the problem of low accuracy and slow recognition speed of traffic signs in specific scenes,a T-YOLO algorithm for traffic sign image recognition is proposed based on the edge information of traffic signs and convolutional neural network(CNN).The algorithm is based on the detection idea of YOLOv2 algorithm,which integrates residual network and convolutional layer with 0 structure.At the same time,the down sampling discards the pooling layer and uses the convolutional layer instead of the pooling layer.Edge information is extracted and the upper sampling is used to improve the accuracy.A 7-layer feature extraction network is designed to shorten the recognition speed.Then,the Softmax function is used to realize multi-classification,batch normalization and multi-scale training are used to shorten the training time.The experimental results show that the algorithm is real and effective.On graphic processing unit(GPU) platform,〖JP2〗the fastest detection speed is 13.69 ms/frame(shorten 9.51 ms per frame),the highest average accuracy rate is 97.3%(improved 7.1%),which meets the requirements of real-time and high-precision recognition.Compared with other algorithms,this algorithm has greatly improved the speed and accuracy of traffic sign recognition,which makes it more suitable for real scenes and closer to vehicle embedded systems. |
| Key words: intelligent transportation system vehicle embedded system traffic sign recognition deep learning T-YOLO multi-scale training residual network |
