基于Mask R-CNN的无人机影像路面交通标志检测与识别

doi:10.6046/gtzyyg.2020.04.09

摘要
图/表
参考文献
相关文章
Metrics

全文: PDF(4341 KB)

HTML
输出: BibTeX | EndNote (RIS)

摘要

交通标志的检测与识别是智能驾驶导航系统的重要组成部分,但传统方法的处理过程由于精度低、时间复杂度高以及鲁棒性差等缺点,不能满足当前智能驾驶的需求。为此,提出了一种基于Mask R-CNN的无人机影像路面交通标志检测与识别方法。首先,制作了一套高质量的无人机影像路面交通标志数据集; 然后,根据统计的200个标记路标特征,对Mask R-CNN中区域候选网络(region proposal network,RPN)结构的锚框宽高比及初始参数进行了改进,使其更好地应用于无人机影像路标场景; 最后,采用精确度-召回率(precision-recall,PR)曲线和平均精度值(mean average precision,mAP)进行精度评价。实验结果表明,锚框宽高比为1∶1,1∶2,1∶3时效果更好; 该方法得到的识别结果平均检测精度为98.33%,高于Faster R-CNN和YOLOv3方法,具有较好的有效性。

	服务

	把本文推荐给朋友
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	陈朋弟
	黄亮
	夏炎
	余晓娜
	高霞霞

关键词 ：无人机影像, 交通标志检测, 交通标志识别, Mask R-CNN, RPN, 锚框

Abstract：

The detection and recognition of traffic signs is an important part of the intelligent driving navigation system. However, due to the shortcomings of low accuracy, high time complexity and poor robustness, the traditional method cannot meet the current needs of intelligent driving. Therefore, a method for detecting and recognizing road traffic signs of UAV images based on Mask R-CNN is proposed. Firstly, a set of high-quality UAV images road traffic sign data sets are produced. Then, based on the statistics of 200 labeled landmarks features, the region proposal network (RPN) structure anchor boxes width-to-height ratio and initial parameters in Mask R-CNN are improved to make it better applied to UAV images road sign scenes. Finally, the precision-recall (PR) curve and mean average precision (mAP) are used for accuracy evaluation. The experimental results show that the anchor boxes width-to-height ratio is better when the ratio is 1∶1, 1∶2, 1∶3; and that the average detection accuracy obtained by this method is 98.33%, which is higher than the accuracy of Faster R-CNN and YOLOv3, indicating better effectiveness.

Key words： UAV images traffic sign detection traffic sign recognition Mask R-CNN RPN anchor boxes

收稿日期: 2020-01-15 出版日期: 2020-12-23

P231

基金资助:国家自然科学基金项目“南方山地城镇建设用地与变化的坡度梯度效应研究”(41961039);云南省应用基础研究计划面上项目“基于全卷积神经网络的多源遥感影像变化检测”(2018FB078);云南省高校工程中心建设计划项目;自然资源部地球观测与时空信息科学重点实验室项目“基于直觉模糊集理论的多源遥感影像变化检测方法研究”(201911);昆明理工大学学生课外学术科技创新基金项目“基于Mask Grid R-CNN的无人机影像路面交通标志检测与识别系统”(2020YB002)

通讯作者: 黄亮

作者简介: 陈朋弟(1993-),男,硕士研究生,主要研究方向为目标检测与遥感影像分割。Email:cpdhn1058475189@163.com。

引用本文:

陈朋弟, 黄亮, 夏炎, 余晓娜, 高霞霞. 基于Mask R-CNN的无人机影像路面交通标志检测与识别[J]. 国土资源遥感, 2020, 32(4): 61-67.
CHEN Pengdi, HUANG Liang, XIA Yan, YU Xiaona, GAO Xiaxia. Detection and recognition of road traffic signs in UAV images based on Mask R-CNN. Remote Sensing for Land & Resources, 2020, 32(4): 61-67.

链接本文:

https://www.gtzyyg.com/CN/10.6046/gtzyyg.2020.04.09 或 https://www.gtzyyg.com/CN/Y2020/V32/I4/61

Tab.1 示例数据集

Fig.1 Mask R-CNN网络框架

Fig.2 FPN结构

Fig.3 RPN网络结构

Fig.4 RoI Align双线性插值

Fig.5 Mask R-CNN的损失率

Fig.6 Mask R-CNN的PR曲线

Tab.2 示例图统计

Tab.3 3种方法的检测结果

Tab.4 Mask R-CNN检测结果

[1]	Chira I M, Chibulcutean A, Danescu R G. Real-time detection of road markings for driving assistance applications[C]// International Conference on Computer Engineering and Systems. IEEE, 2010:158-163.
[2]	Schreiber M, Poggenhans F, Stiller C. Detecting symbols on road surface for mapping and localization using OCR[C]// IEEE International Conference on Intelligent Transportation Systems. IEEE, 2014:597-602.
[3]	Wu T, Ranganathan A. A practical system for road marking detection and recognition[C]// Intelligent Vehicles Symposium. IEEE, 2012,7(2272):25-30.
[4]	Maier G, Pangerl S, Schindler A. Real-time detection and classification of arrow markings using curve-based prototype fitting[C]// Intelligent Vehicles Symposium. IEEE, 2011:442-447.
[5]	Ding D, Yoo J, Jekyo J, et al. Efficient road-sign detection based on machine learning[J]. Bulletin of Networking Computing Systems and Software, 2015,4(1):15-17.
[6]	胡忠闯. 基于深度学习的车道线检测算法研究[D]. 杭州:浙江大学, 2018.
	Hu Z C. Deep learning based lane markings detection algorithm[D]. Hangzhou:Zhejiang University, 2018.
[7]	任洪梅. 基于深度学习的路面交通标志识别[J]. 信息通信, 2017(4):26-27.
	Ren H M. Road traffic sign recognition based on deep learning[J]. Information and Communications, 2017(4):26-27.
[8]	张永涛. 基于机器视觉的路面交通标志识别的应用研究[D]. 广州:广东工业大学, 2015.
	Zhang Y T. Research on the application of pavement traffic signs recognition based on machine vision[D]. Guangzhou:Guangdong University of Technology, 2015.
[9]	Danescu R, Nedevschi S. Detection and classification of painted road objects for intersection assistance applications[C]// International IEEE Conference on Intelligent Transportation Systems. IEEE, 2010:433-438.
[10]	Wang N, Liu W, Zhang C M, et al. The detection and recognition of arrow markings recognition based on monocular vision[C]// IEEE International Symposium on Industrial Electronics. IEEE, 2009:4380-4386.
[11]	Liu W, Lyu J, Yu B, et al. Multi-type road marking recognition using Adaboost detection and extreme learning machine classification[C]// Intelligent Vehicles Symposium (IV). IEEE, 2015:41-46.
[12]	李强. 基于多层次融合与卷积神经网络的道路交通标线检测与识别[D]. 西安:长安大学, 2018.
	Li Q. Road traffic marking and recognition based on multi-level fusion and convolutional neural networks[D]. Xi’an:Chang’an University, 2018.
[13]	Husan V, Hyung H, Jin K, et al. Recognition of damaged arrow-road markings by visible light camera sensor based on convolutional neural network[J]. Sensors, 2016,16(12):2160.
[14]	Yang T, Long X, Sangaiah A K, et al. Deep detection network for real-life traffic sign in vehicular networks[J]. Computer Networks, 2018,136:95-104.
[15]	谢锦, 蔡自兴, 邓海涛, 等. 基于图像不变特征深度学习的交通标志分类[J]. 计算机辅助设计与图形学学报, 2017,29(4):632-640.
	Xie J, Cai Z X, Deng H T, et al. Traffic sign classification based on deep learning of image invariant feature[J]. Journal of Computer-Aided Design and Computer Graphics, 2017,29(4):632-640.
[16]	Balducci F, Impedovo D, Pirlo G. Detection and validation of tow-away road sign licenses through deep learning methods[J]. Sensors, 2018,18(12):4147.
[17]	Filatov D M, Ignatiev K V, Serykh E V. Neural network system of traffic signs recognition[C]// IEEE International Conference on Soft Computing and Measurements (SCM). IEEE, 2017:1488-1491.
[18]	杨梦梦. 基于Faster R-CNN的交通标志检测方法[D]. 武汉:湖北工业大学, 2019.
	Yang M M. Method for traffic sign detection based on Faster R-CNN[D]. Wuhan:Hubei University of Technology, 2019.
[19]	何洪亮. 基于卷积神经网络的交通标志检测研究[D]. 南京:南京理工大学, 2018.
	He H L. Traffic sign detection based on convolutional neural network[D]. Nanjing:Nanjing University of Science and Technology, 2018.
[20]	Zhu Y, Zhang C, Zhou D, et al. Traffic sign detection and recognition using fully convolutional network guided proposals[J]. Neurocomputing, 2016,214:758-766.
[21]	伍锡如, 雪刚刚. 基于图像聚类的交通标志CNN快速识别算法[J]. 智能系统学报, 2019,14(4):670-678.
	Wu X R, Xue G G. CNN-based image clustering algorithm for fast recognition of traffic signs[J]. CAAI Transactions on Intelligent Systems, 2019,14(4):670-678.
[22]	卢飞宇. 基于改进的卷积神经网络的交通标志检测与识别算法[J]. 工业控制计算机, 2019,32(6):54-56.
	Lu F Y. Traffic sign detection and recognition algorithm based on improved multi-task convolutional neural network[J]. Industrial Control Computer, 2019,32(6):54-56.
[23]	Kryvinska N, Poniszewska-Maranda A, Greguš M. An approach towards service system building for road traffic signs detection and recognition[J]. Procedia Computer Science, 2018,141:64-71.
[24]	He K M, Georgia G, Piotr D, et al. Mask R-CNN[C]// Proceedings of the IEEE Computer Society Conference on Computer Vision, 2017:2961-2969.
[25]	Ren S, He K M, Girshick R, et al. Faster R-CNN:Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015,39(6):1137-1149. pmid: 27295650
[26]	Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[C]// Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2015:3431-3440.
[27]	Lin T Y, Piotr D, Ross G, et al. Feature pyramid networks for object detection[C]// Proceedings of the IEEE Computer Society Conference on Computer Vision. IEEE, 2017:2117-2125.
[28]	石杰, 周亚丽, 张奇志. 基于改进Mask R-CNN和Kinect的服务机器人物品识别系统[J]. 仪器仪表学报, 2019,40(4):216-228.
	Shi J, Zhou Y L, Zhang Q Z. Service robot item recognition system based on improved Mask R-CNN and Kinect[J]. Chinese Journal of Scientific Instrument, 2019,40(4):216-228.
[29]	He K M, Zhang X Y, Ren S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[C]// European Conference on Computer Vision.Springer, 2014:346-361.
[30]	Redmon J, Farhadi A. YOLOv3:An incremental improvement[EB/OL]. (2018-04-08)[2020-01-07]. https://arxiv.org/abs/1804.02767.

[1]	蔡祥, 李琦, 罗言, 齐建东. 面向对象结合深度学习方法的矿区地物提取[J]. 国土资源遥感, 2021, 33(1): 63-71.
[2]	夏炎, 黄亮, 陈朋弟. 模糊超像素分割算法的无人机影像烟株精细提取[J]. 国土资源遥感, 2021, 33(1): 115-122.
[3]	王伟, 王新盛, 姚婵, 金添, 邬佳昱, 苏伟. 基于无人机影像的小麦植株密度估算方法研究[J]. 国土资源遥感, 2020, 32(4): 111-119.
[4]	连会青, 孟璐, 韩瑞刚, 杨艺, 余标. 基于无人机遥感的地质信息提取——以柳江盆地为例[J]. 国土资源遥感, 2020, 32(3): 136-142.
[5]	薛武, 赵玲, 于英. 无人机影像稳健区域网平差[J]. 国土资源遥感, 2020, 32(2): 40-45.
[6]	娄佩卿, 陈晓雨, 王疏桐, 付波霖, 黄永怡, 唐廷元, 凌铭. 基于无人机影像的喀斯特农耕区地物识别——以桂林市为例[J]. 国土资源遥感, 2020, 32(1): 216-223.
[7]	李朋龙, 丁忆, 胡艳, 罗鼎, 段松江, 舒文强. 一种基于GPU并行计算的无人机影像快速镶嵌方法[J]. 国土资源遥感, 2017, 29(4): 57-63.
[8]	王旭东, 段福洲, 屈新原, 李丹, 余攀锋. 面向对象和SVM结合的无人机数据建筑物提取[J]. 国土资源遥感, 2017, 29(1): 97-103.
[9]	耿娟, 何成龙, 刘宪鑫. 基于CSIFT特性的无人机影像匹配[J]. 国土资源遥感, 2016, 28(1): 93-100.
[10]	李政, 李永树, 楚彬, 唐敏. 基于加权整体最小二乘法的无人机影像配准[J]. 国土资源遥感, 2014, 26(2): 69-73.
[11]	唐敏, 李永树, 李歆, 刘波. 无人机影像局部增强方法及其在影像匹配中的应用[J]. 国土资源遥感, 2013, 25(4): 53-57.
[12]	何少林, 徐京华, 张帅毅. 面向对象的多尺度无人机影像土地利用信息提取[J]. 国土资源遥感, 2013, 25(2): 107-112.
[13]	鲁恒, 李永树, 何敬. 大重叠度无人机影像自动展绘控制点方法研究[J]. 国土资源遥感, 2011, 23(4): 69-73.

Viewed

Full text

Abstract

Cited

Shared

Discussed