基于改进的ViBe和YOLO v3算法的行人检测方法

李士骥; 李忠民; 李威

基于改进的ViBe和YOLO v3算法的行人检测方法

南昌航空大学信息工程学院, 江西南昌 330063

基金项目:

国家自然科学基金 61263040

江西省自然科学基金 20202BABL202005

南昌航空大学“三小”项目 2022XG13

详细信息

作者简介:
李士骥（1998-），男，硕士研究生，主要从事图像处理方面的研究。E-mail：sdlishiji@163.com

通讯作者:
李忠民（1975-），男，博士，副教授，硕士生导师，主要从事图像处理与人工智能方面的研究。E-mail：zhongmli@163.com

中图分类号: TP391
计量
- 文章访问数: 157
- HTML全文浏览量: 49
- PDF下载量: 39
- 被引次数: 0
出版历程
- 收稿日期: 2022-03-18
- 修回日期: 2022-04-04
- 刊出日期: 2023-02-20

Pedestrian Detection Method Based on Improved ViBe and YOLO v3 Algorithms

School of Information Engineering, Nanchang Hangkong University, Nanchang 330063, China

摘要

摘要: 针对传统视觉背景提取（visual background extractor，ViBe）算法在进行行人检测时会产生鬼影的缺点，本文提出了一种基于改进的ViBe和YOLO v3算法的行人检测方法。利用改进的YOLO v3算法YOLO v3-SPP（spatial pyramid pooling）对ViBe算法的初始化策略进行改进以消除鬼影。运用YOLO v3-SPP算法对首帧图像进行行人检测，使用本文提出的行人消除方法将检测出的行人进行消除，并将输出图像代替ViBe算法的首帧，从而达到消除鬼影的目的。经过分析和实验验证，结果表明该算法能够有效解决鬼影问题。
- 图像处理 /
- 行人检测 /
- 视觉背景提取 /
- 鬼影消除 /
- YOLO v3算法
Abstract: This paper presents an improved algorithm to solve the problem of ghosting in the visual background extractor (ViBe) algorithm for pedestrian detection. The initialization strategy of the ViBe algorithm is improved using the YOLO v3-spatial pyramid pooling (SPP) algorithm to eliminate ghosts. Thus, the YOLO v3-SPP algorithm detects pedestrians in the first frame, eliminates the detected pedestrians, and replaces the first frame of the ViBe algorithm with the output image to eliminate ghosts. The results of the analysis and verification show that the algorithm can effectively solve the ghost problem.
- image processing /
- pedestrian detection /
- visual background extraction /
- ghost elimination /
- YOLO v3

HTML全文

图 1 当前帧像素值与样本值在二维色彩空间的比较

Figure 1. Comparison of current frame pixel value and sample value in two-dimensional color space

下载: 全尺寸图片幻灯片

图 2 SPP层结构

Figure 2. SPP layer structure

下载: 全尺寸图片幻灯片

图 3 YOLO v3-SPP结构

Figure 3. YOLO v3-SPP structure

下载: 全尺寸图片幻灯片

图 4 鬼影问题展示

Figure 4. Ghost problem display

下载: 全尺寸图片幻灯片

图 5 基于YOLO v3 SPP的改进ViBe算法框图

Figure 5. Block diagram of improved ViBe algorithm based on YOLO v3 SPP

下载: 全尺寸图片幻灯片

图 6 分成六部分的行人

Figure 6. Pedestrians divided into six parts

下载: 全尺寸图片幻灯片

图 7 像素代替方法的总流程

Figure 7. General flow chart of pixel replacement method

下载: 全尺寸图片幻灯片

图 8 行人消除的效果

Figure 8. Effect of pedestrian elimination

下载: 全尺寸图片幻灯片

图 9 鬼影消除的效果

Figure 9. Results of ghost elimination

下载: 全尺寸图片幻灯片

图 10 PETS2006视频检测效果：(a)(b)(c)原始图像；(d)(e)(f)ViBe算法；(g)(h)(i)三帧差分法；(j)(k)(l)高斯混合模型算法；(m)(n)(o)本文算法

Figure 10. PETS2006 video detection effect: (a)(b)(c) Original image; (d)(e)(f) ViBe algorithm; (g)(h)(i) Three-frame difference method; (j)(k)(l) Gaussian mixture model algorithm; (m)(n)(o) The algorithm in our paper

下载: 全尺寸图片幻灯片

表 1 实验结果评价

Table 1. Evaluation of experimental results

Algorithm	Precision	Recall	F1-score	Frame per second
ViBe algorithm	0.7186	0.4535	0.5560	5.2629
Three-frame differential method	0.5354	0.6422	0.5839	9.9624
Gaussian mixture model algorithm	0.6552	0.3554	0.4617	4.1905
Proposed algorithm	0.8056	0.7070	0.7531	5.3419

下载: 导出CSV

参考文献(17)

[1]	Sobral A, Vacavant A. A comprehensive review of background subtraction algorithms evaluated with synthetic and real videos[J]. Computer Vision and Image Understanding, 2014, 122(5): 4-21.
[2]	Barnich O, Van Droogenbroeck M. ViBe: a universal background subtraction algorithm for video sequences[J]. IEEE Transactions on Image Processing, 2011, 20(6): 1709-1724. doi: 10.1109/TIP.2010.2101613
[3]	闫硕, 陈科山. 基于双背景模型的鬼影抑制方法研究[J]. 计算机应用与软件, 2016, 33(5): 162-165. doi: 10.3969/j.issn.1000-386x.2016.05.040 YAN Shuo, CHEN Keshan. Study on ghost suppression method based on double background model[J]. Computer Applications and Software, 2016, 33(5): 162-165. doi: 10.3969/j.issn.1000-386x.2016.05.040
[4]	郭冲宇, 杜慧敏. 视觉背景提取算法关键技术研究[J]. 计算机应用研究, 2017, 34(5): 1548-1552, 1589. doi: 10.3969/j.issn.1001-3695.2017.05.061 GUO Chongyu, DU Huimin. Key technologies of visual background extraction algorithm[J]. Computer Application Research, 2017, 34(5): 1548-1552, 1589. doi: 10.3969/j.issn.1001-3695.2017.05.061
[5]	陈树, 丁保阔. 一种基于改进视觉背景提取算法的前景检测[J]. 计算机工程与科学, 2018, 40(4): 673-680. doi: 10.3969/j.issn.1007-130X.2018.04.014 CHEN Shu, DING Baokuo. A foreground detection based on improved visual background extraction algorithm[J]. Computer Engineering and Science, 2018, 40(4): 673-680. doi: 10.3969/j.issn.1007-130X.2018.04.014
[6]	HAN X, GAO Y, LU Z, et al. Research on moving object detection algorithm based on improved three frame difference method and optical flow[C]//Fifth International Conference on Instrumentation & Measurement, 2015: 580-584.
[7]	杨丹, 戴芳. 运动目标检测的ViBe算法改进[J]. 中国图象图形学报, 2018, 23(12): 1813-1828. doi: 10.11834/jig.180304 YANG Dan, DAI Fang. Improvement of ViBe algorithm for moving target detection[J]. China Journal of Image Graphics, 2018, 23(12): 1813-1828. doi: 10.11834/jig.180304
[8]	方岚, 于凤芹. 去除鬼影及阴影的视觉背景提取运动目标检测算法[J]. 激光与光电子学进展, 2019, 56(13): 131002. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ201913009.htm FANG Lan, YU Fengqin. Moving object detection algorithm based on removed ghost and shadow visual background extractor[J]. Laser & Optoelectronics Progress, 2019, 56(13): 131002. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ201913009.htm
[9]	ZHAO D W, TAN J J, YANG W B, et al. An improved vibe algorithm for fast suppression of ghosts and static objects[C]//IEEE International Conference on Mechatronics and Automation(ICMA), 2018: 889-893.
[10]	Girshick R. Fast R-CNN[C]//IEEE International Conference on Computer Vision, 2015: 1440-1448.
[11]	Redmon J Farhadi A. YOLO v3 anincremental improvement[EB/OL]. [2018-04-08] [2022-02-08]. https://arxiv.org/abs/1804.02767.
[12]	HUANG G, LIU Z, Weinberger K Q, et al. Densely connected convolutional networks[C]//IEEE Conference on Computer Vision and Pattern Recognition, 2017: 2261-2269.
[13]	HE K M, ZHANG X Y, REN S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual Recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9): 1904-1916. doi: 10.1109/TPAMI.2015.2389824
[14]	PETS2006. http://www.cvg.reading.ac.uk/PETS2006/data.html [DB/OL]. 2006.
[15]	赵建. 基于三帧差法的运动目标检测方法研究[D]. 西安: 西安电子科技大学, 2013. ZHAO Jian. The Research of Moving Target Detection Method Based on Three-frame Difference[D]. Xi'an: Xidian University, 2013.
[16]	STAUFFER C, GRIM SON W. Learning patterns of activityusing real-time tracking[J]. IEEE Translations on Pattern Analysis and Machine Intelligence, 2000, 22(8): 747-757. doi: 10.1109/34.868677
[17]	Goyette N, Jodoin P M, Porikli F, et al. Change detection. net: a new change detection benchmark dataset[C]//Computer Vision and Pattern Recognition Workshops, 2012: 1-8.