南京大学学报(自然科学版) ›› 2019, Vol. 55 ›› Issue (2): 231–237.doi: 10.13232/j.cnki.jnju.2019.02.008

• • 上一篇    下一篇

一种基于Faster-RCNN的车辆实时检测改进算法

杨 薇,王洪元*,张 继,张中宝   

  1. 常州大学信息科学与工程学院,常州,213164
  • 接受日期:2018-11-21 出版日期:2019-04-01 发布日期:2019-03-31
  • 通讯作者: 王洪元 E-mail:hywang@cczu.edu.cn
  • 基金资助:
    国家自然科学基金(61572085)

An improved vehicle real-time detection algorithm based on Faster-RCNN

Yang Wei,Wang Hongyuan*,Zhang Ji,Zhang Zhongbao   

  1. School of Information Science and Engineering,Changzhou University,Changzhou,213164,China
  • Accepted:2018-11-21 Online:2019-04-01 Published:2019-03-31
  • Contact: Wang Hongyuan E-mail:hywang@cczu.edu.cn

摘要: 随着交通愈加发达,道路愈加拥堵,如何实时准确地获取车辆基本信息以便交通部门及时管理特定路段和路口的车辆显得日益重要. 对交通视频中车辆的检测和识别,不仅需要实时检测,还要保证其准确性. 针对实际情况中车辆之间的遮挡、光照的变化、阴影、道路旁树枝的晃动、背景中固定对象的移动等因素严重影响检测与识别的精度的问题,提出基于Faster-RCNN(Faster-Regions with CNN features)的车辆实时检测改进算法. 首先采用k-means算法对KITTI数据集的目标框进行聚类,得到合适的长宽比,并增加一组尺度(642)以适应差异较大的车辆尺寸;然后改进区域提案网络,降低计算量,优化网络结构;最后在训练阶段采用多尺度策略,降低漏检率,提高精确率. 实验结果表明:改进后的车辆检测算法的mAP(mean Average Precision)达到了82.20%,检测速率为每张照片耗时0.03875 s,基本能够满足车辆实时检测的需求.

关键词: 车辆实时检测, Faster-RCNN, k-means算法, 区域提案网络, 多尺度训练

Abstract: With the development of traffic and road congestion,real-time and accurate access to basic vehicle information makes it increasingly important for the traffic department to manage vehicles on specific sections and intersections in time. In order to detect and identify vehicles in traffic video,the system not only needs real-time detection,but also guarantees the accuracy. In the actual situation,due to the factors which seriously affect the accuracy of detection and recognition,such as the occlusion between vehicles,the change of illumination,shadow,the shaking of branches near the road and the movement of fixed objects in the background,an improved vehicle real-time detection based on Faster-Regions with CNN features(Faster-RCNN)is proposed. First of all,in order to obtain a proper length-width ratio,the k-means algorithm is used to cluster the bounding boxes of the KITTI dataset. At the same time,the algorithm adds a set of scales(642)to adapt to the different vehicle size. Then the RPN(Region Proposal Network)is improved to reduce the amount of calculations and optimize the network structure. Finally,a multi-scale strategy was adopted during the training phase to reduce the missed detection rate and increase the precision. The experimental results show that the mAP(mean Average Precision)of the improved vehicle detection algorithm is 81.83% and the detection speed is 0.03877 seconds per sheet,which can meet the real-time detection requirements of vehicles.

Key words: vehicle real-time detection, Faster-RCNN, k-means algorithm, regional proposal network, multi-scale training

中图分类号: 

  • TP391.41
[1] 周志华. 机器学习. 北京:清华大学出版社,2016,425.(Zhou Z H. Machine learning. Beijing:Tsinghua University Press,2016,425.)
[2] Gerlach M. On the peripheral point spectrum and the asymptotic behavior of irreducible semigroups of Harris operators. Positivity,2013,17(3):875-898.
[3] Zhang C,Ma Y Q. Ensemble machine learning. New York:Springer-Verlag,2012,332.
[4] Dalal N,Triggs B. Histograms of oriented gradients for human detection ∥ IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Diego,CA,USA:IEEE,2005:886-893.
[5] Wang L P. Support vector machines:Theory and applications. Springer Berlin Heidelberg,2005,431.
  [6] Felzenszwalb P F,Girshick R B,McAllester D,et al. Object detection with discriminatively trained part-based models. IEEE Transactions on Pattern Analysis and Machine Intelligence,2010,32(9):1627-1645.
[7] Girshick R,Donahue J,Darrell T,et al. Rich feature hierarchies for accurate object detection and semantic segmentation ∥ The IEEE Conference on Computer Vision and Pattern Recognition. Columbus,OH,USA:IEEE,2014:580-587.
[8] He K M,Zhang X Y,Ren S Q,et al. Spatial pyramid pooling in deep convolutional networks for visual recognition. IEEE Transactions on Pattern Analysis & Machine Intelligence,2015,37(9):1904-1916.
[9] Girshick R. Fast R-CNN ∥ IEEE International Conference on Computer Vision. Santiago,Chile:IEEE,2015:1440-1448.
[10] Ren S Q,He K M,Girshick R,et al. Faster R-CNN:Towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis & Machine Intelligence,2017,39(6):1137-1149.
[11] Redmon J,Divvala S,Girshick R,et al. You only look once:Unified,real-time object detection ∥ IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas,NV,USA:IEEE,2016:779-788.
[12] Liu W,Anguelov D,Erhan D,et al. SSD:Single shot MultiBox detector ∥ European Conference on Computer Vision. Amsterdam,the Netherlands:Springer Cham,2016:21-37.
[13] Redmon J,Farhadi A. YOLO9000:Better,faster,stronger ∥ 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu,HI,USA:IEEE,2017:6517-6525.
[14] Uijlings J R R,van de Sande K E A,Gevers T,et al. Selective search for object recognition. International Journal of Computer Vision,2013,104(2):154-171.
[15] Huang Z X. Extensions to the k-means algorithm for clustering large data sets with categorical values. Data Mining & Knowledge Discovery,1998,2(3):283-304.
[16] Fan Q F,Brown L,Smith J. A closer look at Faster R-CNN for vehicle detection ∥ 2016 IEEE Intelligent Vehicles Symposium(IV). Gothen-burg,Sweden:IEEE,2016:124-129.
[1] 陈睿, 伏云发. 基于EEG握力变化及想象单次识别研究[J]. 南京大学学报(自然科学版), 2020, 56(2): 159-166.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
[1] 林 銮,陆武萍,唐朝生,赵红崴,冷 挺,李胜杰. 基于计算机图像处理技术的松散砂性土微观结构定量分析方法[J]. 南京大学学报(自然科学版), 2018, 54(6): 1064 -1074 .
[2] 马宏亮, 万建武, 王洪元. 一种嵌入样本流形结构与标记相关性的多标记降维算法[J]. 南京大学学报(自然科学版), 2019, 55(1): 92 -101 .
[3] 李嘉明, 邹 勇, 郑 浩, 魏钟波, 杨柳燕, 缪爱军. 养殖塘生态系统重金属污染状况与风险评价[J]. 南京大学学报(自然科学版), 2019, 55(2): 272 -281 .
[4] 王冬丽,申俊峰,邱海成,杜佰松,李建平,聂潇,王业晗. 辽宁五龙金矿黄铁矿标型特征研究及深部找矿预测[J]. 南京大学学报(自然科学版), 2019, 55(6): 898 -915 .
[5] 段友祥,柳璠,孙歧峰,李洪强. 基于相带划分的孔隙度预测[J]. 南京大学学报(自然科学版), 2019, 55(6): 934 -941 .
[6] 陈超逸,林耀进,唐莉,王晨曦. 基于邻域交互增益信息的多标记流特征选择算法[J]. 南京大学学报(自然科学版), 2020, 56(1): 30 -40 .
[7] 游杰, 胡广, 张玺华, 沈安江, 彭瀚霖, 田兴旺, 赵东方. 微生物碳酸盐岩同生⁃早成岩阶段有机质降解示踪:以四川盆地灯影组四段为例[J]. 南京大学学报(自然科学版), 2020, 56(3): 308 -321 .
[8] 戴海亮,沈斌,李开开,张小涛,徐学敏,许智超,周晶晶. 地质条件约束下川北二叠系大隆组富有机质页岩热模拟生烃过程及特征研究[J]. 南京大学学报(自然科学版), 2020, 56(3): 382 -392 .
[9] 刘鑫,胡军,张清华. 属性组序下基于代价敏感的约简方法[J]. 南京大学学报(自然科学版), 2020, 56(4): 469 -479 .