一种改进的适用于监控视频的轻量级入侵检测算法及其应用

陈涛; 陈天宇; 万永菁; 王嵘; 孙静

doi:10.14135/j.cnki.1006-3080.20201110002

一种改进的适用于监控视频的轻量级入侵检测算法及其应用

doi: 10.14135/j.cnki.1006-3080.20201110002

陈涛^1,,
陈天宇¹,
万永菁^1, ,,
王嵘¹,
孙静²

1.
华东理工大学信息科学与工程学院，上海 200237
2.
上海卓希智能科技有限公司研发部，上海 201611

基金项目: 国家自然科学基金（61872143）

详细信息

作者简介:
陈涛：陈　涛(1996—)，男，江苏泰州人，硕士生，主要研究方向为目标检测、深度学习。E-mail：ctzj1026@163.com

通讯作者:
万永菁，E-mail：wanyongjing@ecust.edu.cn

中图分类号: TP391.4
计量
- 文章访问数: 269
- HTML全文浏览量: 160
- PDF下载量: 5
- 被引次数: 0
出版历程
- 收稿日期: 2020-11-10
- 网络出版日期: 2021-01-25

An Improved Lightweight Intrusion Detection Algorithm and Its Application

1.
School of Information Science and Engineering, East China University of Science and Technology, Shanghai 200237, China
2.
Department of Research and Development, Shanghai Joosee Smart Technology Company Limited, Shanghai 201611, China

摘要

摘要: 由于传统的目标检测算法较为复杂，在算力、存储空间有限的场景下无法实时检测，因此本文提出了一种轻量级入侵检测算法。首先采用自适应更新率的混合高斯前景提取算法提取初筛目标，然后基于改进的残差压缩网络(R-SqueezeNet)对初筛目标进行识别分类。实验结果表明，该算法在不降低检测精度的前提下，比传统算法的检测速度平均提升了30倍，模型体积缩减至YOLOv3-tiny算法的1/40。
- 监控视频 /
- 入侵检测 /
- 轻量级 /
- 自适应更新率 /
- R-SqueezeNet
Abstract: With the development of target detection algorithms, intrusion detection based on surveillance video has attracted more and more attention. At present, the traditional target detection algorithm is more complex, and can not be detected in real time in the scene of limited computing power and storage space.Aiming at this pain point, a lightweight intrusion detection algorithm is proposed: firstly extract the preliminary screening target through the adaptive update rate of the mixed Gaussian foreground extraction algorithm, and then identify the preliminary screening target based on the improved residual squeeze network (R-SqueezeNet) classification. Experimental results show that, without reducing the detection accuracy, the algorithm can increase the detection speed by an average of 30 times compared with the traditional algorithm, and the model size is reduced to 1/40 of YOLOv3-tiny.
- surveillance video /
- intrusion detection /
- lightweight /
- adaptive update rate /
- R-SqueezeNet

HTML全文

图 1 整体算法流程

Figure 1. Overall algorithm flow

下载: 全尺寸图片幻灯片

图 2 镜头突变前后的提取效果对比

Figure 2. Comparison of extraction effect before and after lens mutation

下载: 全尺寸图片幻灯片

图 3 不同$ {t}_{H} $下算法的表现

Figure 3. Algorithm performance under different $ {t}_{H} $

下载: 全尺寸图片幻灯片

图 4 提取效果对比

Figure 4. Comparison of extraction effect

下载: 全尺寸图片幻灯片

图 5 R-SqueezeNet结构

Figure 5. Structure of R-SqueezeNet

下载: 全尺寸图片幻灯片

图 6 Fire module计算流程

Figure 6. Calculation process of fire module

下载: 全尺寸图片幻灯片

图 7 残差结构

Figure 7. Residual structure

下载: 全尺寸图片幻灯片

图 8 不同前景调整方式的分类精度对比

Figure 8. Comparison of different foreground resizing method

下载: 全尺寸图片幻灯片

图 9 不同$ {t}_{H} $下的算法性能对比

Figure 9. Performance comparison of algorithms at different $ {t}_{H} $

下载: 全尺寸图片幻灯片

表 1 分类网络模型对比

Table 1. Comparison of classification network models

Model	Number of Fire module	Accuracy/%			Inference time/ms	Size/MB
Model	Number of Fire module	This paper	cats vs dogs	cifar-10	Inference time/ms	Size/MB
SqueezeNet	1	90.56	89.9	91.6	1.34	0.16
	2	94.35	90.61	93.85	1.7	0.33
	3	96.32	92.96	95.74	2.3	0.89
	4	96.58	93.17	96.13	3.78	1.5
	5	96.67	93.39	96.37	4.04	2.8
R-SqueezeNet	3	96.55	93.2	96.1	2.88	0.89
	4	96.73	93.32	96.35	3.92	1.5
	5	96.8	93.49	96.55	4.56	2.8
	6	97.01	93.56	96.61	7.02	4.1
	7	97.2	93.71	96.69	7.68	6.4
	8	97.38	93.87	96.73	9.3	8.9

下载: 导出CSV

表 2 基于自适应和非自适应前景提取的算法对比

Table 2. Comparison of algorithms based on adaptive and non-adaptive foreground extraction

Based	Backbone	FD/%	MD/%
Non-adaptive extraction	LeNet	13.8	8.2
	AlexNet	9.1	3.3
	ZFNet	8.7	2.9
	R-SqueezeNet	8.6	2.7
Adaptive extraction	LeNet	10.3	8.2
	AlexNet	5.4	3.3
	ZFNet	5.1	2.9
	R-SqueezeNet	4.9	2.7

下载: 导出CSV

表 3 本文算法和传统目标检测算法对比

Table 3. Comparison with traditional object detection algorithm

Algorithm	Backbone	Size/MB	FD/%	MD/%	FPS
SSD^[6]	VGG16^[25]	95.7	5.3	2.9	1
RetinaNet^[7]	ResNet50^[19]	146.1	4.2	2.4	<1
YOLOv2^[8]	Darknet19^[8]	194	4.5	2.5	<1
YOLOv3^[9]	Darknet53^[9]	246.9	4.2	2.3	1
YOLOv3-tiny^[9]	Darknet13^[9]	35.6	7.6	3.7	5
This paper	R-SqueezeNet	0.89	4.9	2.7	44

下载: 导出CSV

参考文献(25)

[1]	赵潇. 基于人类视觉系统的监控视频目标提取技术研究[D]. 重庆: 重庆邮电大学, 2018.
[2]	DALAL N, TRIGGS B. Histograms of oriented gradients for human detection[C]//2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. USA: IEEE, 2005: 886-893.
[3]	VIOLA P, JONES M. Rapid object detection using a boosted cascade of simple features[C]//Proceedings the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. USA: IEEE, 2001: 511-518.
[4]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. USA: IEEE, 2014: 580-587.
[5]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. USA: IEEE, 2016: 779-788.
[6]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: Single shot multibox detector[C]//European Conference on Computer Vision. Cham: Springer, 2016: 21-37.
[7]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE International Conference on Computer Vision. Italy: IEEE, 2017: 2980-2988.
[8]	REDMON J, FARHADI A. YOLO9000: Better, faster, stronger[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. USA: IEEE, 2017: 7263-7271.
[9]	REDMON J, FARHADI A. Yolov3: An incremental improvement[EB/OL]. arxiv. org, (2018-04-10)[2020-11-01]. https://arxiv.org/pdf/1804.02767.pdf.
[10]	IANDOLA F N, HAN S, MOSKEWICZ M W, et al. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and< 0.5 MB model size[EB/OL]. arxiv. org, (2016-03-20)[2020-11-01]. https://arxiv.org/pdf/1602.07360v3.pdf.
[11]	STAUFFER C, GRIMSON W E L. Learning patterns of activity using real-time tracking[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 22(8): 747-757. doi: 10.1109/34.868677
[12]	ZIVKOVIC Z. Improved adaptive Gaussian mixture model for background subtraction[C]//Proceedings of the 17th International Conference on Pattern Recognition. UK: IEEE, 2004: 28-31.
[13]	ZIVKOVIC Z, VAN DER HEIJDEN F. Efficient adaptive density estimation per image pixel for the task of background subtraction[J]. Pattern Recognition Letters, 2006, 27(7): 773-780. doi: 10.1016/j.patrec.2005.11.005
[14]	LEE D S. Effective Gaussian mixture learning for video background subtraction[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2005, 27(5): 827-832. doi: 10.1109/TPAMI.2005.102
[15]	龚安, 牛博, 史海涛. 基于分块的帧差法和混合高斯算法的油田作业区入侵检测[J]. 计算机与数字工程, 2019, 47(12): 3041-3044.
[16]	刘馨. 监控视频中的图像颜色评价与优化[D]. 杭州: 浙江大学, 2015.
[17]	李均, 王志诚, 吴雨轩, 等. 熵概念的延拓——从热熵到信息熵[J]. 大学物理, 2020, 39(10): 29-33.
[18]	王林, 王超凡. 差分信息熵在拼接图像质量评估中的应用[J]. 计算机仿真, 2020, 37(4): 265-268, 273. doi: 10.3969/j.issn.1006-9348.2020.04.055
[19]	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. USA: IEEE, 2016: 770-778.
[20]	汪斌, 陈宁. 基于残差注意力U-Net结构的端到端歌声分离模型[J]. 华东理工大学学报(自然科学版). doi: 10.14135/j.cnki.1006-3080.20200903001.
[21]	高磊, 范冰冰, 黄穗. 基于残差的改进卷积神经网络图像分类算法[J]. 计算机系统应用, 2019, 28(7): 139-144.
[22]	LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324. doi: 10.1109/5.726791
[23]	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90. doi: 10.1145/3065386
[24]	ZEILER M D, FERGUS R. Visualizing and understanding convolutional networks[C]//European Conference on Computer Vision. Cham: Springer, 2014: 818-833.
[25]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. arxiv. org, (2014-10-22)[2020-11-01]. https://arxiv.org/abs/1409.1556.