高级搜索

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于改进YOLOv4-tiny的轻量化室内人员目标检测算法

赵凤 李永恒 李晶 刘汉强

赵凤, 李永恒, 李晶, 刘汉强. 基于改进YOLOv4-tiny的轻量化室内人员目标检测算法[J]. 电子与信息学报, 2022, 44(11): 3815-3824. doi: 10.11999/JEIT220241
引用本文: 赵凤, 李永恒, 李晶, 刘汉强. 基于改进YOLOv4-tiny的轻量化室内人员目标检测算法[J]. 电子与信息学报, 2022, 44(11): 3815-3824. doi: 10.11999/JEIT220241
ZHAO Feng, LI Yongheng, LI Jing, LIU Hanqiang. Lightweight Indoor Personnel Detection Algorithm Based on Improved YOLOv4-tiny[J]. Journal of Electronics & Information Technology, 2022, 44(11): 3815-3824. doi: 10.11999/JEIT220241
Citation: ZHAO Feng, LI Yongheng, LI Jing, LIU Hanqiang. Lightweight Indoor Personnel Detection Algorithm Based on Improved YOLOv4-tiny[J]. Journal of Electronics & Information Technology, 2022, 44(11): 3815-3824. doi: 10.11999/JEIT220241

基于改进YOLOv4-tiny的轻量化室内人员目标检测算法

doi: 10.11999/JEIT220241
基金项目: 国家自然科学基金(62071379, 62071378, 61901365, 62106196),陕西省自然科学基础研究计划 (2021JM-461, 2020JM-299),西安邮电大学西邮新星团队资助项目(xyt2016-01)
详细信息
    作者简介:

    赵凤:女,教授,研究方向为智能信息处理、模式识别与图像处理

    李永恒:男,硕士生,研究方向为深度学习与目标检测

    李晶:男,高级工程师,研究方向为智能信息处理

    刘汉强:男,副教授,研究方向为模式识别与图像处理

    通讯作者:

    赵凤 fzhao.xupt@gmail.com

  • 中图分类号: TN911.73

Lightweight Indoor Personnel Detection Algorithm Based on Improved YOLOv4-tiny

Funds: The National Natural Science Foundation of China (62071379, 62071378, 61901365, 62106196), The Natural Science Basic Research Plan in Shaanxi Province of China (2021JM-461, 2020JM-299), Funded Project of New Star Team of Xi'an University of Posts & Telecommunications (xyt2016-01)
  • 摘要: 深度学习在室内人员检测领域应用广泛,但是传统的卷积神经网络复杂度大且需要高算力GPU的支持,很难实现在嵌入式设备上的部署。针对上述问题,该文提出一种基于改进YOLOv4-tiny的轻量化室内人员目标检测算法。首先,设计一种改进的Ghost卷积特征提取模块,有效减少了模型的复杂度;同时,该文通过采用带有通道混洗机制的深度可分离卷积进一步减少网络参数;其次,该文构建了一种多尺度空洞卷积模块以获得更多具有判别性的特征信息,并结合改进的空洞空间金字塔池化结构和具有位置信息的注意力机制进行有效的特征融合,在提升准确率的同时提高推理速度。在多个数据集和多种硬件平台上的实验表明,该文算法在精度、速度、模型参数和体积等方面优于原YOLOv4-tiny网络,更适合部署于资源有限的嵌入式设备。
  • 图  1  Ghost卷积

    图  2  改进YOLOv4-tiny的轻量化室内人员检测网络结构图

    图  3  Ghost 卷积特征提取网络模块

    图  4  多尺度空洞卷积融合模块结构图

    图  5  基于通道混洗机制的深度可分离卷积模块

    图  6  改进的ASPP结构

    图  7  Coordinate Attention模块结构图

    图  8  YOLOv4-tiny与本文算法在不同场景下检测效果对比图

    图  9  多场景下多指标综合对比图

    表  1  不同扩张率下实验结果

    多尺度空洞卷积融合模块空洞空间金字塔池化特征融合模块
    扩张率精确率(%)召回率(%)mAP(%)扩张率精确率(%)召回率(%)mAP(%)
    [2,2,2]79.6968.2981.75[3,3,3]79.4978.9782.91
    [4,4,4]79.5567.9581.15[9,9,9]78.9579.2182.49
    [2,3,4]76.1880.4382.93[2,4,6]80.3278.5682.66
    [3,2,4]79.3378.9682.52[3,6,9]76.1880.4382.93
    [4,5,6]79.3578.6282.12[12,14,18]79.4677.6682.46
    下载: 导出CSV

    表  2  模块验证结果

    ghost blockCBLCSASPPCAdilated conv block参数量(M)FLOPs(G)模型体积(MB)精确率(%)召回率(%)mAP(%)
    模型A1.230.925.580.7657.9275.28
    模型B1.221.025.679.6862.7477.14
    模型C1.441.056.381.3464.2879.33
    模型D1.441.056.580.0969.1981.13
    模型E1.611.466.476.1880.4382.93
    下载: 导出CSV

    表  3  多个数据集下检测效果对比(%)

    数据集名称评价指标YOLOv4-tiny本文算法
    PASCAL VOC Person数据集精确率76.7376.18
    召回率62.8380.43
    mAP74.6382.93
    INRIA数据集精确率90.8198.13
    召回率75.0079.23
    mAP88.8691.74
    CUHK Occlusion 数据集精确率90.9789.71
    召回率73.8572.82
    mAP82.4786.03
    机房环境自建数据集精确率74.6895.82
    召回率96.3188.36
    mAP95.7293.84
    下载: 导出CSV

    表  4  不同网络模型结果对比

    模型类型模型名称参数量(M)FLOPs(G)模型体积(MB)精确率(%)召回率(%)mAP(%)
    通用目标检测网络YOLOv4[10]64.3630.16277.776.2184.5386.63
    SSD[3]26.1559.5290.769.3771.1872.15
    EfficientDet[4]3.872.5514.979.8470.8282.17
    轻量化网络YOLOv4-tiny[9]5.913.4322.576.7362.8374.63
    MobileNet-SSDv2[22]6.071.5514.576.3164.5575.86
    YOLOv4-MobileNet v1[23]12.264.9851.475.1280.2681.96
    YOLOv4-MobileNet v2[24]10.373.7846.875.9780.0082.96
    YOLOv4-MobileNet v3[25]11.303.5154.170.9773.8582.47
    YOLOv4-GhostNet[26]11.003.2542.777.4578.0183.10
    本文算法1.611.466.476.1880.4382.93
    下载: 导出CSV

    表  5  不同性能设备推理速度对比

    模型类型模型名称fps(帧/s)帧图片推理耗时(ms)
    GPU环境
    RTX2070
    CPU环境
    I5-8200U
    Jetson NxJetson NanoGPU环境
    RTX2070
    CPU环境
    I5-8200U
    Jetson NxJetson Nano
    通用目标检测网络YOLOv4[10]260.025.171.463849710193680
    SSD[3]690.3510.802.86142853917349
    EfficientDet[4]180.144.803.46547022207288
    轻量化网络YOLOv4-tiny[9]1014.0124.0012.489.902494080
    Mobilenet-SSDv2[22]762.3319.0014.471342550469
    YOLOv4-MobileNet v1[23]501.2015.305.031982765198
    YOLOv4-MobileNet v2[24]441.1713.205.252284975190
    YOLOv4-MobileNet v3[25]371.2611.905.512679283181
    YOLOv4-GhostNet[26]301.279.704.2033786102238
    本文算法1059.0127.0016.019.521153762
    下载: 导出CSV
  • [1] 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, 2014: 580–587.
    [2] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 779–788.
    [3] LIU Wei, ANGUELOV D, ERHAN D, et al. SSD: Single shot MultiBox detector[C]. The 14th European Conference on Computer Vision, Amsterdam, The Netherlands, 2016: 21–37.
    [4] TAN Mingxing, PANG Ruoming, and LE Q V. Efficientdet: Scalable and efficient object detection[C]. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, USA, 2020: 10778–10787.
    [5] LIU Wei, LIAO Shengcai, HU Weidong, et al. Learning efficient single-stage pedestrian detectors by asymptotic localization fitting[C]. The 15th European Conference on Computer Vision, Munich, Germany, 2018: 643–659.
    [6] 张明伟, 蔡坚勇, 李科, 等. 基于DE-YOLO的室内人员检测方法[J]. 计算机系统应用, 2020, 29(1): 203–208. doi: 10.15888/j.cnki.csa.007240

    ZHANG Mingwei, CAI Jianyong, LI Ke, et al. Indoor personnels detection method based on DE-YOLO[J]. Computer Systems &Applications, 2020, 29(1): 203–208. doi: 10.15888/j.cnki.csa.007240
    [7] 董小伟, 韩悦, 张正, 等. 基于多尺度加权特征融合网络的地铁行人目标检测算法[J]. 电子与信息学报, 2021, 43(7): 2113–2120. doi: 10.11999/JEIT200450

    DONG Xiaowei, HAN Yue, ZHANG Zheng, et al. Metro pedestrian detection algorithm based on multi-scale weighted feature fusion network[J]. Journal of Electronics &Information Technology, 2021, 43(7): 2113–2120. doi: 10.11999/JEIT200450
    [8] 苏杨, 卢翔, 李琨, 等. 基于轻量深度学习网络的机房人物检测研究[J]. 工业仪表与自动化装置, 2021(1): 100–103. doi: 10.3969/j.issn.1000-0682.2021.01.024

    SU Yang, LU Xiang, LI Kun, et al. Research on computer room human detection based on lightweight deep learning network[J]. Industrial Instrumentation &Automation, 2021(1): 100–103. doi: 10.3969/j.issn.1000-0682.2021.01.024
    [9] WANG C Y, BOCHKOVSKIY A, and LIAO H Y M. Scaled-YOLOv4: Scaling cross stage partial network[C]. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Nashville, USA, 2021: 13024–13033.
    [10] BOCHKOVSKIY A, WANG C Y, and LIAO H Y M. YOLOv4: Optimal speed and accuracy of object detection[EB/OL]. https://arxiv.org/abs/2004.10934v1, 2020.
    [11] HAN Kai, WANG Yunhe, TIAN Qi, et al. GhostNet: More features from cheap operations[C]. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, USA, 2020: 1577–1586.
    [12] ZHANG Xiangyu, ZHOU Xinyu, LIN Mengxiao, et al. ShuffleNet: An extremely efficient convolutional neural network for mobile devices[C]. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 6848–6856.
    [13] YU F, KOLTUN V, and FUNKHOUSER T. Dilated residual networks[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 636–644.
    [14] CHEN L C, PAPANDREOU G, KOKKINOS I, et al. DeepLab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 40(4): 834–848. doi: 10.1109/TPAMI.2017.2699184
    [15] HOU Qibin, ZHOU Daquan, and FENG Jiashi. Coordinate attention for efficient mobile network design[C]. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Nashville, USA, 2021: 13708–13717.
    [16] SANDLER M, HOWARD A, ZHU Menglong, et al. MobileNetV2: Inverted residuals and linear bottlenecks[C]. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 4510–4520.
    [17] HE Kaiming, ZHANG Xiangyu, REN Shaoqing, 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
    [18] GHIASI G, LIN T Y, and LE Q V. DropBlock: A regularization method for convolutional networks[C]. The 32nd International Conference on Neural Information Processing Systems, Montréal, Canada, 2018: 10750–10760.
    [19] EVERINGHAM M, VAN GOOL L, WILLIAMS C K I, et al. The PASCAL visual object classes (VOC) challenge[J]. International Journal of Computer Vision, 2010, 88(2): 303–338. doi: 10.1007/s11263-009-0275-4
    [20] DALAL N and TRIGGS B. Histograms of oriented gradients for human detection[C]. 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, San Diego, USA, 2005: 886–893.
    [21] OUYANG Wanli and WANG Xiaogang. A discriminative deep model for pedestrian detection with occlusion handling[C]. 2012 IEEE Conference on Computer Vision and Pattern Recognition, Providence, USA, 2012: 3258–3265.
    [22] CHIU Y C, TSAI C Y, RUAN M D, et al. Mobilenet-SSDv2: An improved object detection model for embedded systems[C]. 2020 International Conference on System Science and Engineering, Kagawa, Japan, 2020: 1–5.
    [23] LIU Jie and LIU Lizhi. Helmet wearing detection based on YOLOv4-MT[C]. Proceedings of the 2021 4th International Conference on Robotics, Control and Automation Engineering, Wuhan, China, 2021: 1–5.
    [24] FANG Lifa, WU Yanqiang, LI Yuhua, et al. Ginger seeding detection and shoot orientation discrimination using an improved YOLOv4-LITE network[J]. Agronomy, 2021, 11(11): 2328. doi: 10.3390/agronomy11112328
    [25] WANG Shengying, CHEN Tao, LV Xinyu, et al. Forest fire detection based on lightweight Yolo[C]. The 2021 33rd Chinese Control and Decision Conference, Kunming, China, 2021: 1560–1565.
    [26] WANG Huixuan, GE Huayong, and LI Muxian. PFG-YOLO: A safety helmet detection based on YOLOv4[C]. The 2021 IEEE 5th Information Technology, Networking, Electronic and Automation Control Conference, Xi'an, China, 2021: 1242–1246.
  • 加载中
图(9) / 表(5)
计量
  • 文章访问数:  552
  • HTML全文浏览量:  520
  • PDF下载量:  209
  • 被引次数: 0
出版历程
  • 收稿日期:  2022-03-08
  • 修回日期:  2022-06-28
  • 网络出版日期:  2022-07-05
  • 刊出日期:  2022-11-14

目录

    /

    返回文章
    返回