[1]李奇武,杨小军.基于改进 YOLOv4 的轻量级车辆检测方法[J].计算机技术与发展,2023,33(01):42-48.[doi:10. 3969 / j. issn. 1673-629X. 2023. 01. 007]
　LI Qi-wu,YANG Xiao-jun.Lightweight Vehicle Detection Method Based on Improved YOLOv4[J].,2023,33(01):42-48.[doi:10. 3969 / j. issn. 1673-629X. 2023. 01. 007]
点击复制

基于改进 YOLOv4 的轻量级车辆检测方法()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
33
期数:
2023年01期
页码:
42-48
栏目:
媒体计算
出版日期:
2023-01-10

文章信息/Info

Title:
Lightweight Vehicle Detection Method Based on Improved YOLOv4
文章编号:
1673-629X(2023)01-0042-07
作者:
李奇武; 杨小军
长安大学 信息工程学院,陕西 西安 710064
Author(s):
LI Qi-wu; YANG Xiao-jun
School of Information Engineering,Chang’an University,Xi’an 710064,China
关键词:
车辆检测; 轻量化网络; YOLOv4; MobileNetV3; CBAM
Keywords:
vehicle detection; lightweight network; YOLOv4; MobileNetV3; CBAM
分类号:
TP399
DOI:
10. 3969 / j. issn. 1673-629X. 2023. 01. 007
摘要:
道路车辆检测是智能交通管控的重要组成部分。 针对现有车辆检测算法模型容量大、参数数量较多、占用内存多,难以在智能交通监控场景中适用于算力和内存均有限的边缘设备的问题,提出一种改进 YOLOv4 的轻量化车辆检测方法 MC-YOLO。 为了减少算法的参数量,把 YOLOv4 网络模型压缩到合适的大小,对网络的部分结构做了针对性的设计:保留原 YOLOv4 的主干网以外的其余模块,将 YOLOv4 的主干 CSPDarknet53 替换为 MobileNetV3 使其轻量化;另外,为了弥补主干网络轻量化导致的车辆检测效果的下降,在检测网络的骨干和加强特征提取部分之间插入 CBAM 模块,提升车辆检测模型的性能。 实验结果表明,经过改进的车辆检测算法在 UA-DETRAC 数据集上体现了良好的性能,平均精度与原YOLOv4 算法相近,模型在参数量上比原 YOLOv4 模型减少了约 77% ,模型大小仅为 55. 3 MB,较原 YOLOv4 模型减少了约190 MB。 改进后的车辆检测算法在模型轻量化的同时不仅能够保证较高的检测精度,而且能够满足在算力资源有限的边缘设备进行实时性车辆检测的需求。
Abstract:
Road vehicle detection is an important part of intelligent traffic control. In response to the problem that existing algorithms forvehicle detection tend to have a relatively large model capacity,　a large number of parameters and a large memory footprint,making itdifficult to apply to edge devices with limited arithmetic power and memory in intelligent traffic monitoring scenarios, a lightweightvehicle detection method MC-YOLO with improved YOLOv4 is proposed. In order to reduce the number of algorithm parameters andcompress the YOLOv4 network model to a suitable size,a targeted design was made for part of the network structure. The remainingmodules other than the original YOLOv4 backbone were retained, and the YOLOv4 backbone CSPDarknet53 was replaced byMobileNetV3 to make it lighter. In addition,to compensate for the degradation of vehicle detection due to the lighter backbone network,aCBAM module was inserted between the backbone and the enhanced feature extraction part of the detection network,and this attentionmodule was used to improve the performance of the vehicle detection model. The experimental results demonstrate that the improvedvehicle detection algorithm performs well on the UA-DETRAC dataset,with an average accuracy similar to that of the original YOLOv4algorithm,and the model size is only 55. 3 MB,which is about 190 MB less than that of the original YOLOv4 model,with about 77%less number of parameters. The improved vehicle detection algorithm can ensure high detection accuracy while lightweight the model,which can meet the demand for real-time vehicle detection in edge devices with limited computing power resources.

相似文献/References:

[1]方宏 杜正春.车辆检测线远程监控系统的研制[J].计算机技术与发展,2009,(12):185.
　FANG Hong,DU Zheng-chun.Development of Remote Monitoring System of Vehicle Inspection Lane[J].,2009,(01):185.
[2]张明恒 王华莹 郭烈.基于改进K—Means算法的车辆识别方法[J].计算机技术与发展,2012,(05):53.
　ZHANG Ming-heng,WANG Hua-ying,GUO Lie.Method of Vehicle Detection Based on Improved K-Means Algorithm[J].,2012,(01):53.
[3]吴玉枝,吴志红,熊运余.基于卷积神经网络的小样本车辆检测与识别[J].计算机技术与发展,2018,28(06):1.[doi:10.3969/ j. issn.1673-629X.2018.06.001]
　WU Yu-zhi,WU Zhi-hong,XIONG Yun-yu.Vehicle Detection and Recognition of a Few Samples Based on Convolutional Neural Network[J].,2018,28(01):1.[doi:10.3969/ j. issn.1673-629X.2018.06.001]
[4]王传钦[],曹江涛[],姬晓飞[]. 基于视频分析技术的车距测量及预警系统设计[J].计算机技术与发展,2016,26(09):87.
　WANG Chuan-qin[],CAO Jiang-tao[],JI Xiao-fei[]. Design of a Vehicle Distance Measurement and Early Warning System Based on Video Analysis Techniques[J].,2016,26(01):87.
[5]阮航,孙涵.基于 Faster R-CNN 的车辆多属性识别[J].计算机技术与发展,2018,28(10):129.[doi:10.3969/ j. issn.1673-629X.2018.10.027]
　RUAN Hang,SUN Han.Vehicle Multi-attribute Recognition Based on Faster R-CNN[J].,2018,28(01):129.[doi:10.3969/ j. issn.1673-629X.2018.10.027]
[6]杨岳航,孙涵.基于部件模型的复杂场景车辆检测方法[J].计算机技术与发展,2018,28(12):34.[doi:10．3969/j． issn．1673－629X．2018．12．007]
　YANG Yuehang,SUN Han.A Vehicle Detection Method Based on Part Model in Complex Traffic Scenes[J].,2018,28(01):34.[doi:10．3969/j． issn．1673－629X．2018．12．007]
[7]阮航,王立春.基于特征图的车辆检测和分类[J].计算机技术与发展,2018,28(11):39.[doi:10.3969/ j. issn.1673-629X.2018.11.009]
　RUAN Hang,WANG Li-chun.Vehicle Detection and Classification Based on Feature Map[J].,2018,28(01):39.[doi:10.3969/ j. issn.1673-629X.2018.11.009]
[8]陶晓力,刘宁钟,沈家全.基于深度信息融合的航拍车辆检测[J].计算机技术与发展,2019,29(09):117.[doi:10. 3969 / j. issn. 1673-629X. 2019. 09. 023]
　TAO Xiao-li,LIU Ning-zhong,SHEN Jia-quan.Aerial Vehicle Detection Based on Depth Information Fusion[J].,2019,29(01):117.[doi:10. 3969 / j. issn. 1673-629X. 2019. 09. 023]
[9]王 翎,孙 涵.基于域适应的多场景车辆检测[J].计算机技术与发展,2019,29(12):158.[doi:10. 3969 / j. issn. 1673-629X. 2019. 12. 028]
　WANG Ling,SUN Han.Vehicle Detection in Different Scenes Based on Domain Adaptation[J].,2019,29(01):158.[doi:10. 3969 / j. issn. 1673-629X. 2019. 12. 028]
[10]余明高,王连涛,闵凡蕾.面向智能交通引导的车辆检测算法改进[J].计算机技术与发展,2022,32(09):43.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 007]
　YU Ming-gao,WANG Lian-tao,MIN Fan-lei.An Improved Vehicle Detection Algorithm for Intelligent Traffic Guidance[J].,2022,32(01):43.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 007]

更新日期/Last Update: 2023-01-10