基于两阶段分类算法的中国交通标志牌识别

doi:10.19562/j.chinasae.qcgc.2022.03.016

摘要/Abstract

摘要：

自动驾驶技术对于缓解交通拥堵，降低交通运输成本具有重要作用；高级驾驶辅助系统（ADAS）可以有效增加汽车驾驶的舒适性和安全性。交通标志牌中包含了丰富的语义信息，为自动驾驶汽车和ADAS的决策提供重要约束，因此交通标志牌的识别算法开发至关重要。本文基于中国交通场景特点以及自动驾驶、ADAS对于交通标志牌识别的高准确性需求，提出了一种基于两阶段分类的交通标志牌识别算法框架。算法包含检测和分类两个阶段，检测阶段检测出图像中的交通标志牌，分类阶段对交通标志牌先后进行大类和子类划分。算法通过细化任务，独立提升各算法模块的性能，进而提高整体算法的识别精度。本文对单阶段识别算法进行改进作为算法的检测模块，实验结果表明，提出的算法精度上优于基准单阶段识别算法，mAP平均提升8.52%，并且在检测速度优于传统两阶段识别算法Faster RCNN的情况下，mAP提升40%以上。

关键词: 目标识别, 交通标志牌识别, 卷积神经网络

Abstract:

Autonomous driving technology plays an important role in alleviating traffic congestion and reducing transportation cost. Advanced driver assistance systems （ADAS） can effectively increase the comfort and safety of car driving. Traffic signs contain rich semantic information， which provides important constraints for decision-making of autonomous vehicles and ADAS. Therefore， development of traffic sign recognition algorithms is very important. Based on the characteristics of traffic scenes in China and the high accuracy requirements of automatic driving and ADAS on traffic sign detection， this paper proposes a traffic sign recognition algorithm framework based on two-stage classification. The algorithm consists of two stages of recognition and classification. In the recognition stage， traffic signs in the image are detected. And in the classification stage， traffic signs are divided into categories and subcategories. By refining the task， the algorithm improves the performance of each algorithm module independently， thus improving the recognition accuracy of the whole algorithm. In this paper， the single-stage recognition algorithm is improved to be used as the recognition module of the algorithm. The experimental results show that the proposed algorithm is better than the benchmark single-stage recognition algorithm in accuracy， with an average increase of 8.52% in mAP. In addition， with the detection speed better than the traditional two-stage recognition algorithm Faster-RCNN， the mAP is improved by 40%.

Key words: object recognition, traffic sign recognition, convolutional neural network

冯润泽,江昆,于伟光,杨殿阁. 基于两阶段分类算法的中国交通标志牌识别[J]. 汽车工程, 2022, 44(3): 434-441.

Runze Feng,Kun Jiang,Weiguang Yu,Diange Yang. Chinese Traffic Sign Recognition Based on Two-stage Classification Algorithm[J]. Automotive Engineering, 2022, 44(3): 434-441.

图/表 12

表 1

图 1

图 2

图 3

表 2

表 3

图 4

图 5

表 4

表 5

图 6

图 7

参考文献 32

1	YANG D G， JIANG K， ZHAO D， et al. Intelligent and connected vehicles： current status and future perspectives［J］. Science China（Technological Sciences）， 2018， 61（10）：1446-1471.
2	FU M Y， HUANG Y S. A survey of traffic sign recognition［C］. 2010 International Conference on Wavelet Analysis and Pattern Recognition. IEEE， 2010： 119-124.
3	AKATSUKA H， IMAI S. Road signposts recognition system［C］. SAE Paper 870239.
4	DE LA ESCALERA A， MORENO L E， SALICHS M A， et al. Road traffic sign detection and classification［J］. IEEE Transactions on Industrial Electronics， 1997， 44（6）： 848-859.
5	BROGGI A， CERRI P， MEDICI P， et al. Real time road signs recognition［C］. 2007 IEEE Intelligent Vehicles Symposium. IEEE， 2007： 981-986.
6	PICCIOLI G， DE MICHELI E， PARODI P， et al. Robust method for road sign detection and recognition［J］. Image and Vision Computing， 1996， 14（3）： 209-223.
7	ALBAWI S， MOHAMMED T A， AL-ZAWI S. Understanding of a convolutional neural network［C］. 2017 International Conference on Engineering and Technology （ICET）. IEEE， 2017： 1-6.
8	ISHAK K A， SANI M M， TAHIR N M， et al. A speed limit sign recognition system using artificial neural network［C］. 2006 4th Student Conference on Research and Development. IEEE， 2006： 127-131.
9	KANG D S， GRISWOLD N C， KEHTARNAVAZ N. An invariant traffic sign recognition system based on sequential color processing and geometrical transformation［C］. Proceedings of the IEEE Southwest Symposium on Image Analysis And Interpretation. IEEE， 1994： 88-93.
10	ESCALERA S， RADEVA P， PUJOL O. Traffic sign classification using error correcting techniques［J］. VISAPP （2）， 2007， 2007： 281-285.
11	GAVRILA D M， PHILOMIN V. Real-time object detection for “smart” vehicles［C］. Proceedings of the Seventh IEEE International Conference on Computer Vision. IEEE， 1999， 1： 87-93.
12	AOYAGI Y， ASAKURA T. A study on traffic sign recognition in scene image using genetic algorithms and neural networks［C］. Proceedings of the 1996 IEEE IECON. 22nd International Conference on Industrial Electronics， Control， and Instrumentation. IEEE， 1996， 3： 1838-1843.
13	ESCALERA A， ARMINGOL J M A， MATA M. Traffic sign recognition and analysis for intelligent vehicles［J］. Image and Vision Computing， 2003， 21（3）：247-258.
14	LOY G， BARNES N. Fast shape-based road sign detection for a driver assistance system［C］. IEEE/RSJ International Conference on Intelligent Robots & Systems. IEEE， 2004.
15	KELLER C G， SPRUNK C， BAHLMANN C， et al. Real-time recognition of U.S. speed signs［C］. IEEE Xplore. IEEE Xplore， 2008.
16	MIURA J， KANDA T， SHIRAI Y. An active vision system for real-time traffic sign recognition［C］. ITSC2000. 2000 IEEE Intelligent Transportation Systems. Proceedings （Cat. No. 00TH8493）. IEEE， 2000： 52-57.
17	LAFUENTE-ARROYO S， GIL-JIMENEZ P， MALDONADO-BASCON R， et al. Traffic sign shape classification evaluation I： SVM using distance to borders［C］. Intelligent Vehicles Symposium， 2005. Proceedings. IEEE. IEEE， 2005.
18	GIL-JIMENEZ P， GOMEZ-MORENO H， SIEGMANN P， et al. Traffic sign shape classification based on support vector machines and the FFT of the signature of blobs［C］. Intelligent Vehicles Symposium. IEEE， 2007.
19	KRIZHEVSKY A， SUTSKEVER I， HINTON G E. Imagenet classification with deep convolutional neural networks［J］. Advances in Neural Information Processing Systems， 2012， 25： 1097-1105.
20	ZOU Z， SHI Z， GUO Y， et al. Object detection in 20 years： A survey［J］. arXiv preprint arXiv：， 2019.
21	YOU S， BI Q， JI Y， et al. Traffic sign detection method based on improved SSD［J］. Information， 2020， 11（10）： 475.
22	GARG P， CHOWDHURY D R， MORE V N. Traffic sign recognition and classification using YOLOv2， Faster RCNN and SSD［C］. 2019 10th International Conference on Computing， Communication and Networking Technologies （ICCCNT）. IEEE， 2019.
23	NGUYEN B T， RYONG S J， KYU K J. Fast traffic sign detection under challenging conditions［C］. 2014 International Conference on Audio， Language and Image Processing. IEEE， 2014： 749-752.
24	ZHANG J， HUANG M， JIN X， et al. A real-time chinese traffic VisioSign detection algorithm based on modified YOLOv2［J］. Algorithms， 2017， 10（4）：127.
25	LIU X F， XIONG F. A real-time traffic sign detection model based on improved YOLOv3［J］. IOP Conference Series Materials Science and Engineering， 2020， 787：012034.
26	王海，王宽，蔡英凤，等. 基于改进级联卷积神经网络的交通标志识别［J］. 汽车工程， 2020， 42（9）： 1256-1262.
	WANG H， WANG K， CAI Y， et al. Traffic sign recognition based on improved cascade convolution neural network［J］. Automotive Engineering， 2020， 42（9）： 1256-1262.
27	MOGELMOSE A， TRIVEDI M M， MOESLUND T B. Vision-based traffic sign detection and analysis for intelligent driver assistance systems： perspectives and survey［J］. IEEE Transactions on Intelligent Transportation Systems， 2012， 13（4）：1484-1497.
28	ZHE Z， LIANG D， ZHANG S， et al. Traffic-sign detection and classification in the wild［C］. 2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. IEEE， 2016.
29	LIN T Y， MAIRE M， BELONGIE S， et al. Microsoft coco： common objects in context［C］. European Conference on Computer Vision. Springer， Cham， 2014： 740-755.
30	ZHENG Z， WANG P， LIU W， et al. Distance-IoU loss： faster and better learning for bounding box regression［C］. Proceedings of the AAAI Conference on Artificial Intelligence. 2020， 34（07）： 12993-13000.
31	TAN M， LE Q. Efficientnet： rethinking model scaling for convolutional neural networks［C］. International Conference on Machine Learning. PMLR， 2019： 6105-6114.
32	SANDLER M， HOWARD A， ZHU M， et al. MobileNetV2： inverted residuals and linear bottlenecks［C］. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. IEEE， 2018.

数据集	年份	图片数	实例数	种类
GTSDB	2013	900	1 200	3
TT100K	2016	10 000	300 00	128
CCTSDB	2017	15 723	21 134	3

模块	操作	层数
1	Conv3×3	1
2	MBConv1， k3×3	2
3	MBConv6， k3×3	3
4	MBConv6， k5×5	3
5	MBConv6， k3×3	6
6	MBConv6， k5×5	6
7	MBConv6， k5×5	6
8	MBConv6， k3×3	1
9	Conv1×1&池化&全连接	1

模型	准确率/%	参数量/10⁶	浮点数运算/10亿
EifficientNet-B3	81.6	12	1.8
ResNet-101	80.9	84	32
PolyNet	81.3	92	35

数据集	图片数	实例数	实例数多于100的种类数
TT-100K	15 723	21 134	45
CCTSDB	10 000	30 000	3
本文数据集	18 955	41 028	53

算法	mAP@0.5/%	准确率/%	召回率/%	误检率/%	检测时长/（ms·帧^-1）	帧率/（帧·s^-1）
Faster RCNN	33.52	89.89	56.79	23.22	74.57	13
YOLOv3	72.05	92.97	75.22	12.09	36.96	27
YOLOv4	72.28	92.36	75.67	11.91	41.38	24
YOLOX	68.13	90.17	65.51	17.42	48.35	20
two-stage-YOLOv3	78.74	94.41	85.34	6.44	28.03+37.13	15
two-stage-YOLOv4	79.83	94.40	84.78	6.23	35.54+37.13	14
Two-stage-YOLOX	79.45	94.80	82.81	6.40	42.36+37.13	13