[1]梁秦嘉,刘 怀,陆 飞.基于改进YOLOv3模型的交通视频目标检测算法研究[J].南京师范大学学报(工程技术版),2021,21(02):047-53.[doi:10.3969/j.issn.1672-1292.2021.02.008]
 Liang Qinjia,Liu Huai,Lu Fei.Traffic Video Target Detection Algorithm Based on Improved YOLOv3[J].Journal of Nanjing Normal University(Engineering and Technology),2021,21(02):047-53.[doi:10.3969/j.issn.1672-1292.2021.02.008]
点击复制

基于改进YOLOv3模型的交通视频目标检测算法研究
分享到:

南京师范大学学报(工程技术版)[ISSN:1006-6977/CN:61-1281/TN]

卷:
21卷
期数:
2021年02期
页码:
047-53
栏目:
计算机科学与技术
出版日期:
2021-06-30

文章信息/Info

Title:
Traffic Video Target Detection Algorithm Based on Improved YOLOv3
文章编号:
1672-1292(2021)02-0047-07
作者:
梁秦嘉刘 怀陆 飞
南京师范大学电气与自动化工程学院,江苏 南京 210023
Author(s):
Liang QinjiaLiu HuaiLu Fei
School of Electrical and Automation Engineering,Nanjing Normal University,Nanjing 210023,China
关键词:
交通监控目标检测深度学习卷积神经网络
Keywords:
traffic monitoringtarget detectiondeep learningconvolutional neural network
分类号:
TP391.4
DOI:
10.3969/j.issn.1672-1292.2021.02.008
文献标志码:
A
摘要:
针对现有的基于YOLOv3的目标检测算法在多尺度目标检测上存在速度与精度难以平衡的问题,在已有算法的基础上改进形成新的YOLOv3多尺度目标检测算法. 该算法首先通过k-means++聚类为各个尺度选择候选锚框的数量和长宽比维数,有效降低原始算法在初始聚类点所造成的聚类偏差; 其次将YOLOv3的检测尺度从3扩展到4,以提高对不同尺度下目标检测的精度; 最后为避免梯度衰落,将检测层前的6个卷积层转换为2个残差单元. 在UA-DETRAC数据集上的实验结果表明,该方法比原始YOLOv3的准确率和召回率分别提高了7.91%和4.57%,同时此算法的处理速度可实现对交通视频的实时处理.
Abstract:
Aiming at the problem that the existing target detection algorithm based on YOLOv3 is difficult to balance the speed and accuracy of multi-scale target detection,the paper proposes an improved YOLOv3 multi-scale target detection algorithm. The algorithm firstly selects the number of candidate anchor frames and the aspect ratio dimensions for each scale through k-means++ clustering,which effectively reduces the clustering deviation caused by the original algorithm at the initial clustering points. Secondly,the YOLOv3 detection scale is extended from 3 to 4,in order to improve the accuracy of target detection under different scales. Finally,in order to avoid gradient fading,the six convolutional layers before the detection layer are converted into two residual units. The experimental results on UA-DETRAC dataset show that the accuracy and recall rate of this method are 7.91% and 4.57% higher than the original YOLOv3,respectively. Meanwhile,the processing speed can meet the real-time requirements of traffic video.

参考文献/References:

[1] 李明熹,林正奎,曲毅. 计算机视觉下的车辆目标检测算法综述[J]. 计算机工程与应用,2019,55(24):20-28.
[2]LECUN Y,BENGIO Y,HINTON G. Deep learning[J]. Nature,2015,521(7553):436-444.
[3]WU L,WANG Y,GAO J,et al. Where-and-when to look:deep siamese attention networks for video-based person re-identification[J]. IEEE Transactions on Multimedia,2019,21(6):1412-1424.
[4]YOLCU G,OZTEL I,KAZAN S,et al. Deep learning-based face analysis system for monitoring customer interest[J]. Journal of Ambient Intelligence and Humanized Computing,2020,11(1):237-248.
[5]崔艳鹏,王元皓,胡建伟. 一种改进YOLOv3的动态小目标检测方法[J]. 西安电子科技大学学报,2020,47(3):1-7.
[6]黄继鹏,史颖欢,高阳. 面向小目标的多尺度Faster-RCNN检测算法[J]. 计算机研究与发展,2019,56(2):319-327.
[7]GIRSHICK R,DONAHUE J,DARRELL T,et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//2014 IEEE Conference on Computer Vision and Pattern Recognition. Columbus,USA:IEEE,2014:580-587.
[8]GIRSHICK R. Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision(ICCV). Santiago,Chile:IEEE,2015:1440-1448.
[9]KOIRALA A,WALSH K B,WANG Z,et al. Deep learning for real-time fruit detection and orchard fruit load estimation:benchmarking of 《MangoYOLO》[J]. Precision Agriculture,2019,20(6):1107-1135.
[10]REN S,HE K,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,2017,39(6):1137-1149.
[11]戴立伟,黄山. 交通场景目标检测指标优化研究[J]. 激光与光电子学进展,2020,57(14):344-352.
[12]LIU W,ANGUELOV D,ERHAN D,et al. SSD:Single shot multiBox detector[C]//European Conference on Computer Vision 2016. Glasgow,UK,2020.
[13]SANG J,WU Z,GUO P,et al. An improved YOLOv2 for vehicle detection[J]. Sensors,2018,18(12):4272.
[14]REDMON J,FARHADI A. YOLOv3:an incremental improvement[EB/OL]. [2020-07-27]. https://arxiv.org/abs/1804.02767.
[15]孔方方,宋蓓蓓. 改进YOLOv3的全景交通监控目标检测[J]. 计算机工程与应用,2020,56(8):20-25.
[16]ZHAO X,PU F,WANG Z,et al. Detection,tracking,and geolocation of moving vehicle from UAV using monocular camera[J]. IEEE Access,2019,7:101160-101170.
[17]XU Z F,JIA R S,SUN H M,et al. Light-YOLOv3:fast method for detecting green mangoes in complex scenes using picking robots[J]. Applied Intelligence,2020,50(12):4670-4687.
[18]HE K,ZHANG X,REN S,et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Las Vegas,USA:IEEE,2016:770-778.
[19]WEN L,DU D,CAI Z,et al. UA-DETRAC:a new benchmark and protocol for multi-object detection and tracking[J]. Computer Science,2020,193:102907.

相似文献/References:

[1]程显毅,胡海涛,季国华,等.基于深度学习监控场景下的多尺度目标检测算法研究[J].南京师范大学学报(工程技术版),2018,18(03):033.[doi:10.3969/j.issn.1672-1292.2018.03.005]
 Cheng Xianyi,Hu Haitao,Ji Guohua,et al.Research on Algorithm of Multi-Scale Target DetectionBased on Deep Learning in Monitoring Scenario[J].Journal of Nanjing Normal University(Engineering and Technology),2018,18(02):033.[doi:10.3969/j.issn.1672-1292.2018.03.005]
[2]梁秦嘉,刘 怀,陆 飞.基于改进YOLOv3的运动目标分类检测算法研究[J].南京师范大学学报(工程技术版),2021,21(04):027.[doi:10.3969/j.issn.1672-1292.2021.04.005]
 Liang Qinjia,Liu Huai,Lu Fei.Moving Target Classification and Detection AlgorithmBased on Improved YOLOv3[J].Journal of Nanjing Normal University(Engineering and Technology),2021,21(02):027.[doi:10.3969/j.issn.1672-1292.2021.04.005]
[3]姜有亮,张锋军,沈沛意,等.基于语义连通图的场景图生成算法[J].南京师范大学学报(工程技术版),2022,22(02):048.[doi:10.3969/j.issn.1672-1292.2022.02.008]
 Jiang Youliang,Zhang Fengjun,Shen Peiyi,et al.Scene Graph Generation Based on Semantic Connected Graph[J].Journal of Nanjing Normal University(Engineering and Technology),2022,22(02):048.[doi:10.3969/j.issn.1672-1292.2022.02.008]
[4]陆 飞,沈世斌,苏晓云,等.基于改进Mask R-CNN的交通监控视频车辆检测算法[J].南京师范大学学报(工程技术版),2020,20(04):044.[doi:10.3969/j.issn.1672-1292.2020.04.007]
 Lu Fei,Shen Shibin,Su Xiaoyun,et al.Vehicle Detection Algorithm Based on Improved Mask R-CNNin Traffic Surveillance Video[J].Journal of Nanjing Normal University(Engineering and Technology),2020,20(02):044.[doi:10.3969/j.issn.1672-1292.2020.04.007]

备注/Memo

备注/Memo:
收稿日期:2020-08-23.
基金项目:国家自然科学基金项目(61603194).
通讯作者:刘怀,博士,副教授,研究方向:数字图像处理、实时控制系统. E-mail:liuhuai@njnu.edu.cn
更新日期/Last Update: 2021-06-30