基于SqueezeNet卷积神经网络的车辆检测

doi:10.11959/j.issn.2096-3750.2020.00175

物联网学报 ›› 2020, Vol. 4 ›› Issue (3): 120-125.doi: 10.11959/j.issn.2096-3750.2020.00175

• 专题：智慧交通物联网 • 上一篇

基于SqueezeNet卷积神经网络的车辆检测

魏泽发¹(),崔华²

¹ 长安大学教育技术与网络中心，陕西西安 710064
² 长安大学信息工程学院，陕西西安 710064

修回日期:2020-07-06 出版日期:2020-09-30 发布日期:2020-09-07
作者简介:魏泽发（1992- ），男，河北衡水人，长安大学助理工程师，主要研究方向为智能交通、深度学习、图像处理等|崔华（1977- ），女，陕西西安人，博士，长安大学教授，主要研究方向为图像分析、数据挖掘、机器学习、深度学习及其在智能交通中的应用研究等
基金资助:
陕西省科技厅重点研发计划(2018ZDXM-GY-047)

Vehicle detection based on SqueezeNet convolutional neural network

Zefa WEI¹(),Hua CUI²

¹ Educational Technology and Network Center,Chang’an University,Xi’an 710064,China
² School of Information Engineering,Chang’an University,Xi’an 710064,China

Revised:2020-07-06 Online:2020-09-30 Published:2020-09-07
Supported by:
The Key R＆D Plan of Shaanxi Provincial Science and Technology Department(2018ZDXM-GY-047)

摘要/Abstract

摘要：

在智能交通系统中，针对车辆目标检测算法可移植性不高、检测速度较慢等问题，提出了一种基于SqueezeNet卷积神经网络的车辆检测方法。通过融合SqueezeNet与SSD（single shot multibox detector）算法的车辆检测方法，在UA-DETRAC数据集上进行训练，实现了车辆目标的快速检测，提升了模型的可移植性，缩短了单帧检测时间。实验结果表明，所提模型在保证准确率的同时，模型单帧检测时间可达22.3 ms，模型大小为16.8 MB，相较于原SSD算法，模型大小减少了约8/9。

关键词: 智能交通, 卷积神经网络, SqueezeNet, 车辆检测

Abstract:

In the intelligent transportation system,aiming at the problem of low portability and speed of detection in vehicle target detection algorithm,a vehicle detection method based on SqueezeNet convolutional neural network was proposed.In order to realize the rapid detection of vehicle targets,improve the portability and shorten the detection time of the single frame,the model was trained on the UA-DETRAC dataset by fusing the SqueezeNet with the single shot multibox detector (SSD) algorithm.The experimental results showed that the time of the single frame detection could reach 22.3 ms and the model size was 16.8 MB.Compared with the original SSD algorithm,the model size was reduced by about 8/9.At the same time,the accuracy of the proposed model was guaranteed.

Key words: intelligent transportation, convolutional neural network, SqueezeNet, vehicle detection

中图分类号:

TP389.1

魏泽发,崔华. 基于SqueezeNet卷积神经网络的车辆检测[J]. 物联网学报, 2020, 4(3): 120-125.

Zefa WEI,Hua CUI. Vehicle detection based on SqueezeNet convolutional neural network[J]. Chinese Journal on Internet of Things, 2020, 4(3): 120-125.

图/表 6

图1

图2

图3

表1

图4

表2

参考文献 18

[1]	DALAL N , TRIGGS B . Histograms of oriented gradients for human detection[J]. IEEE Computer Society, 2015,1: 886-893.
[2]	LOWE D G . Distinctive image features from scale-invariant keypoints[J]. International Journal of Computer Vision, 2004,60(2): 91-110.
[3]	VIOLA P , JONES M . Rapid object detection using a boosted cascade of simple features[J]. Institute of Electrical and Electronics Engineers Computer Society, 2001,1(1): 1511-1518.
[4]	BAY H , ESS A , TUYTELAARS T ,et al. SURF:speeded up robust features[J]. Computer Vision and Image Understanding, 2008,110(3): 346-359.
[5]	SáNCHEZ A V . Advanced support vector machines and kernel methods[J]. Neurocomputing, 2003,55(1): 5-20.
[6]	VIOLA P , JONES M J . Robust real-time face detection[J]. International Journal of Computer Vision, 2004,57(2): 137-154.
[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. IEEE, 2014: 580-587.
[8]	UIJLINGS J R R , SANDE K E A V D , GEVERS T ,et al. Selective search for object recognition[J]. International Journal of Computer Vision, 2013,104(2): 154-171.
[9]	REDMON J , DIVVALA S , GIRSHICK R ,et al. You only look once:unified,real-time object detection[C]// Computer Vision and Pattern Recognition. 2016: 779-788.
[10]	REDMON J , FARHADI A . YOLO9000:better,faster,stronger[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2017: 6517-6525.
[11]	REDMON J , FARHADI A . YOLOv3:an incremental improvement[J]. arXiv:1804.02767, 2018
[12]	LIU W , ANGUELOV D , ERHAN D ,et al. SSD:single shot MultiBox detector[M]. Berlin: SpringerPress, 2016.
[13]	IANDOLA F N , HAN S , MOSKEWICZ M ,et al. SqueezeNet:AlexNet-level accuracy with 50x fewer parameters and < 0.5 MB model size[J]. International Conference on Learning Representations, 2017
[14]	WEN L Y , DU D W , CAI Z W ,et al. UA-DETRAC:a new benchmark and protocol for multi-object detection and tracking[J]. Computer Vision and Image Understanding, 2020,193:102907.
[15]	JIA Y Q , SHELHAMER E , DONAHUE J ,et al. Caffe:convolutional architecture for fast feature embedding[C]// ACM International Conference on Multimedia. ACM, 2014: 675-678.
[16]	ZEILER M D , TAYLOR G W , FERGUS R . Adaptive deconvolutional networks for mid and high level feature learning[C]// 2011 International Conference on Computer Vision. IEEE, 2011: 2018-2025.
[17]	IOFFE S , SZEGEDY C . Batch normalization:accelerating deep network training by reducing internal covariate shift[J]. Journal of Machine Learning Research, 2015: 448-456.
[18]	JISOO J , HYOJIN P , NOJUN K . Enhancement of SSD by concatenating feature maps for object detection[J]. Computer Vision and Pattern Recognition, 2017

模型名称	模型大小/MB	训练时间/h	单张测试时间/ms	准确率
SSD-VGG16	95.5	65	30.6	89.2%
SSD-MobileNet	22.2	57	26.3	88.7%
SSD-ShuffleNet	14.5	36	16.1	82.4%
SSD-SqueezeNet	11.6	42	18.2	88.14%

模型名称	模型大小/MB	训练时间/h	单张测试时间/ms	准确率
SSD-SqueezeNet	11.6	42	18.2	88.14%
SSD-SqueezeNet-R	16.8	50	22.3	90.63%

基于SqueezeNet卷积神经网络的车辆检测

Vehicle detection based on SqueezeNet convolutional neural network

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 6

参考文献 18

相关文章 3

Metrics

推荐阅读 0

[1]	张德干,赵彭真,高瑾馨,张婷,龚倡乐. 面向车联网的智能数据传输新方法[J]. 物联网学报, 2019, 3(2): 89-99.
[2]	余辰,张丽娟,金海. 大数据驱动的智能交通系统研究进展与趋势[J]. 物联网学报, 2018, 2(1): 56-63.
[3]	林尚静,田锦,李锦鸿. 面向公路网智能交通系统应用的无线接入方案评估[J]. 物联网学报, 2017, 1(3): 55-59.