近几年我国机动车保有量急剧增长，交通事故频繁发生。山区公路由于其特殊的行车环境，发生交通事故的致死率高于平原公路和城市道路，而这些事故的发生和驾驶员对交通标志的误判漏判有关。因此对山区公路交通标志识别技术进行研究，对保证驾驶员在山区公路的行车安全、减少山区公路交通事故的发生具有重要现实意义。

本文以山区公路交通标志图像为研究对象，在总结交通标志识别方法和国内外研究现状的基础上，对山区公路交通标志的特征和识别中的问题进行分析研究；考虑到国内山区公路交通标志数据集较为匮乏，通过人工标注制作了山区公路交通标志数据集，并对交通标志进行形态学操作来扩充数据集；由于山区公路的特殊环境，山区公路交通标志图像易受光照和天气因素的干扰，采用改进的Retinex算法来改善交通标志图像的亮度，增强图像的对比度，突出交通标志的特征；针对原YOLOv4模型识别效率低，无法满足山区公路交通标志识别实时性的要求问题，引入了轻量化网络MobileNetv3来作为主干特征提取网络，将PANet中的标准卷积替换为深度可分离卷积，来降低模型的参数量和计算量，并使用K-means++算法对本文数据集中的交通标志进行聚类得到符合当前数据集的先验框尺寸模板，提高了模型的识别效率；为了解决山区公路图像中小尺寸交通标志漏检误检的问题，对YOLOv4的特征金字塔路径聚合网络PANet结构进行改进，将预测网络的尺度由3个增加为四4个，同时设计了模型的定位损失函数，提高了模型整体的识别精度；结合前文交通标志识别模型，根据前后端交互框架完成了山区公路交通标志识别系统的开发，并对系统功能进行了测试，测试结果表明本文方法在准确性和实时性上可以满足山区公路交通标志识别的实际应用要求。

本文通过对山区公路交通标志识别方法的研究，可实现对山区公路场景下交通标志的准确而快速的识别，对保障驾驶员在山区公路的行车安全，推进辅助驾驶系统商业落地有着重要意义，具有一定理论研究意义和工程应用价值。

In recent years, the number of motor vehicles in my country has increased rapidly, and traffic accidents have occurred frequently. Due to its special driving environment, the fatality rate of traffic accidents on mountain roads is higher than that on plain roads and urban roads, and the occurrence of these accidents is related to the misjudgment and omission of traffic signs by drivers. Therefore, it is of great practical significance to study the identification technology of traffic signs on mountain highways to ensure the safety of drivers on mountain highways and to reduce the occurrence of mountain highway traffic accidents.

This paper takes the mountain road traffic sign images as the research object, and on the basis of summarizing and analyzing the traffic sign recognition methods and research status at home and abroad, analyzes and studies the characteristics and problems in the identification of mountain road traffic signs; due to the lack of mountain road traffic sign datasets in China, this paper made a mountain road traffic sign dataset by manual annotation, and performed morphological operations on the traffic signs to expand the dataset; due to the special environment of mountain roads, the traffic sign images of mountain roads are easily disturbed by light and weather factors, this paper uses an improved Retinex algorithm to improve the brightness of traffic sign images, enhance the contrast of images, and highlight the characteristics of traffic signs; in view of the low recognition efficiency of the original YOLOv4 model, which cannot meet the real-time requirements of road traffic sign recognition in mountainous areas, this paper introduces the lightweight network MobileNetv3 as the backbone feature extraction network, and replaces the standard convolution in PANet with a depthwise separable convolution to reduce the amount of parameters and computation of the model, the traffic signs are clustered to obtain a priori frame size template that conforms to the current data set, which improves the recognition efficiency of the model; in order to solve the problem of missed detection and false detection of small-sized traffic signs in mountain highway images, the PANet structure of YOLOv4's feature pyramid path aggregation network is improved, and the scale of the prediction network is increased from 3 to 4, at the same time, the localization loss of the model is designed, which improved the overall recognition accuracy of the model; Combined with the previous traffic sign recognition model , according to the front-end and back-end interaction framework, the development of the mountain highway traffic sign recognition system is completed, and the system function is tested, the test results show that the method in this paper can meet the practical application requirements of mountain highway traffic sign recognition in accuracy and real-time performance.

This paper realizes the accurate and rapid identification of traffic signs in mountain highway scenes through the research on the recognition of traffic signs on mountain highways, which is of great significance to ensure the driving safety of drivers and promote the commercial implementation of assisted driving systems, and has certain theoretical research significance and engineering application value.

[1]中华人民共和国公安部交通管理局https://www.mps.gov.cn/n2254314/n6409334/c8322353/content.html.

[2]中华人民共和国国家统计局.中国统计年鉴2021 http://www.stats.gov.cn/tjsj/ndsj/2021/indexch.htm.

[3]中国交通年鉴2020[M]. 北京: 中国交通年鉴社, 2021.

[4]胡立伟, 薛刚, 李林育, 等. 高原地质及气象环境下公路交通风险致因耦合分析[J]. 中国公路学报, 2018, 31(1): 110-119.

[5]肖润谋, 赵金龙, 陈荫三, 等. 山区公路交通安全标志设计[J]. 长安大学学报(自然科学版), 2006(3): 63-67

[6]陈林. 基于深度学习的路标识别系统研究[D]. 上海: 华东师范大学, 2019.

[7]屈治华, 邵毅明, 邓天民, 等. 复杂光照条件下的交通标志检测与识别[J]. 激光与光电子学进展, 2019, 56(23): 140-147.

[8]Kim J B. Detection of Traffic Signs Based on Eigen-color Model and Saliency Model in Driver Assistance Systems[J]. International Journal of Automotive Technology, 2013, 14(3): 429-439.

[9]Won W J, Lee M, Son J W. Implementation of Road Traffic Signs Detection Based on Saliency Map Model[C]//2008 IEEE Intelligent Vehicles Symposium. IEEE, 2008: 542-547.

[10]吴亥. 基于支持向量机的实时交通标志识别系统[D]. 重庆: 西南大学, 2020.

[11]郝永杰, 周博文. 基于图像特征及改进支持向量机算法的交通标志识别[J]. 现代电子技术, 2017, 40(8): 97-99.

[12]戴雪瑞, 袁雪, 乐国庆, 等. 复杂场景下基于颜色对和MSER的交通标志检测方法[J]. 北京交通大学学报, 2018, 42(5): 107-115.

[13]赵树恩, 刘伟. 基于改进VGG模型的低照度道路交通标志识别[J]. 重庆交通大学学报(自然科学版), 2021, 40(10): 178-184.

[14]Sermanet P, Eigen D, Zhang X, et al. Overfeat: Integrated Recognition, Localization and Detection Using Convolutional Networks[J]. arXiv preprint arXiv: 1312. 6229, 2013.

[15]Yu C S, Shu Q G. Traffic Sign Recognition Based on HOG Feature Extraction[J]. Journal of Measurements in Engineering, 2021, 9(3): 142-155.

[16]Greenhalgh J, Mirmehdi M. Real-Time Detection and Recognition of Road Traffic Signs[J]. IEEE Transactions on Intelligent Transportation Systems, 2012, 13(4): 1498-1506.

[17]Zaklouta F, Stanciulescu B, Hamdoun O. Traffic Sign Classification Using Kd Trees and Random Forests[C]//The 2011 International Joint Conference on Neural Networks. IEEE, 2011: 2151-2155.

[18]詹海浪. 基于计算机视觉的车道线检测与交通路标识别[D]. 广州: 华南理工大学, 2015.

[19]张传伟, 崔万豪. 基于Gabor特征提取和SVM交通标志识别方法研究[J]. 现代电子技术, 2018, 41(17): 136-140.

[20]Sermanet P, Lecun Y. Traffic Sign Recognition with Multi-scale Convolutional Networks[C]//International Joint Conference on Neural Networks. IEEE, 2011: 2809-2813.

[21]Zhu 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: 2110-2118.

[22]Arcos-García Á, Alvarez-Garcia J A, Soria-Morillo L M. Deep Neural Network for Traffic Sign Recognition Systems: An Analysis of Spatial Transformers and Stochastic Optimisation Methods[J]. Neural Networks, 2018, 99: 158-165.

[23]刘琼, 李宗贤, 孙富春, 等. 基于深度信念卷积神经网络的图像识别与分类[J]. 清华大学学报(自然科学版), 2018, 58(9): 781-787.

[24]张兴国, 刘晓磊, 李靖, 等. BP神经网络下的限速交通标志实时检测识别[J]. 西安电子科技大学学报, 2018, 45(5): 136-142.

[25]董娜, 刘欣宇, 吴爱国. 基于人工蜂群的新型圆形交通标志识别算法[J]. 天津大学学报(自然科学与工程技术版), 2019, 52(6): 585-593.

[26]田锋, 雷印杰, 邓棋. 基于YOLOv3的自然路况信息识别研究[J]. 计算机应用研究, 2020, 37(S1): 391-393.

[27]尉天成, 陈小锋, 殷元亮. 基于多尺度卷积神经网络的道路交通标志识别方法研究[J]. 西北工业大学学报, 2021, 39(4): 891-900.

[28]张淑芳, 朱彤. 基于残差单发多框检测器模型的交通标志检测与识别[J]. 浙江大学学报(工学版), 2019, 53(5): 940-949.

[29]徐兢成, 王丽华. 基于AlexNet网络的交通标志识别方法[J]. 无线电工程, 2022, 52(3): 470-475.

[30]孙畅. 基于深度学习的高级辅助驾驶视觉感知关键技术研究[D]. 北京: 北京科技大学, 2022.

[31]秦雅琴, 杨妮, 谢济铭, 等. 山区低等级公路典型路段驾驶人视觉差异性分析[J/OL]. 安全与环境学报: 1-10[2022-04-03].

[32]GB5768.1-2009. 道路交通标志和标线[S]. 北京: 中国标准出版社, 2009.

[33]Lv X, Sun Y, Zhang J, et al. Low-light Image Enhancement via Deep Retinex Decomposition and Bilateral Learning[J]. Signal Processing: Image Communication, 2021, 99: 116466.

[34]Liu K, Liang Y. Dual-purpose Method for De-hazing and Enhancement of Underwater and Low-light Images[J]. Machine Vision and Applications, 2021, 32(5): 1-13.

[35]张航瑛, 王雪琦, 王华英, 等. 基于明度分量的Retinex-Net图像增强改进方法[J/OL]. 物理学报: 2022: 1-15[2022-04-04]. http://kns.cnki.net/kcms/detail/11.1958.O4. 20220302.1838.016.html.

[36]Girshick R, Donahue J, Darrell T, et al. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2014: 580-587.

[37]Girshick R. Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision. 2015: 1440-1448.

[38]Ren S, He K, Girshick R, et al. Faster R-CNN: Towards Realtime Object Detection with Region Proposal Networks[C]//Advances in Neural Information Processing Systems, 2015: 91-99.

[39]Liu W, Anguelov D, Erhan D, et al. SSD: Single Shot Multibox Detector[C]//European conference on computer vision. Springer, Cham, 2016: 21-37.

[40]Redmon J, Divvala S, Girshick R, et al. You Only Look Once: Unified, Real-time Object Detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016: 779-788.

[41]Bochkovskiy A, Wang C Y, Liao H Y M. Yolov4: Optimal Speed and Accuracy of Object Detection[J]. arXiv preprint arXiv:2004.10934, 2020.

[42]Howard A, Sandler M, Chu G, et al. Searching for Mobilenetv3[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019: 1314-1324.

[43]Song S, Que Z, Hou J, et al. An efficient Convolutional Neural Network for Small Traffic Sign Detection[J]. Journal of Systems Architecture, 2019, 97: 269-277.

[44]Liu Z, Li D, Ge S S, et al. Small Traffic Sign Detection from Large Image[J]. Applied Intelligence, 2019, 50(1): 1-13.

[45]梁天骄, 鲍泓, 潘卫国, 等. 基于改进单级特征图方法的交通标志检测[J]. 传感器与微系统, 2022, 41(2): 123-127+131.

[46]刘婷婷, 苗华, 李琳, 等. 融合场景上下文的轻量级目标检测网络[J]. 激光与光电子学进展, 2021, 58(20): 127-135.

[47]王浩, 雷印杰, 陈浩楠. 改进YOLOV3实时交通标志检测算法[J]. 计算机工程与应用, 2022, 58(8): 243-248.

[48]狄岚, 何锐波, 梁久祯. 基于可能性聚类和卷积神经网络的道路交通标识识别算法[J]. 南京大学学报(自然科学), 2019, 55(2): 238-250.

[49]Xu Y, Qu W, Li Z, et al. Efficient K-Means++ approximation with MapReduce[J]. IEEE Transactions on Parallel and Distributed Systems, 2014, 25(12): 3135-3144.

[50]杨文君. K_means算法初始聚类中心点选取方法研究[J]. 科学技术创新, 2019(33): 84-85.

[51]胡伟. 一种改进的K_means聚类方法[J]. 计算机与现代化, 2012(1): 22-24+56.