基于YOLOv5-DeepSORT融合的低帧率水面航行多目标跟踪算法

付诗文, 韩星程, 王黎明, 武国强, 王鸿儒, 马文, 王志勇

doi:10.16411/j.cnki.issn1006-7736.2024.01.011

大连海事大学学报 >

2024 , Vol. 50 >Issue 1: 94 - 101

DOI: https://doi.org/10.16411/j.cnki.issn1006-7736.2024.01.011

基于YOLOv5-DeepSORT融合的低帧率水面航行多目标跟踪算法

展开

(1.中北大学信息与通信工程学院，山西太原 030051；2. 太原重工股份有限公司，山西太原 030024；3. 山西太重数智科技股份有限公司，山西太原 030024；4. 山西国化能源有限责任公司，山西太原 030000)

付诗文（2000 —），男，硕士生，研究方向：图像处理。韩星程（1989 —），男，博士，研究方向：水声信号处理。王黎明*（1976 — ），男，博士，教授，博士生导师，研究方向：电子技术、多维信号处理与重建。武国强（1985 —），男，硕士，研究方向：焦化设备智能化。王鸿儒（1985 —），男，硕士，研究方向：设备智能化。马文（1984 —），男，学士，研究方向：智能检测。王志勇（1984 —），男，硕士，研究方向：天然气管道运行管理。E-mail：wlm@nuc.edu.cn。

收稿日期: 2023-08-21

修回日期: 2023-10-18

录用日期: 2023-10-18

网络出版日期: 2023-08-21

基金资助

国家自然科学青年基金（62203405）； 2021年山西省应用基础研究计划项目（20210302124545）；省部共建动态测试技术国家重点实验室开放研究基金（2022-SYSJJ-08）

收起

A multi-target tracking algorithm for low frame rate surface navigation based on YOLOv5-DeepSORT fusion

Expand

(1.School of Information and Communication Engineering，North University of China, Taiyuan 030051, China; 2. Taiyuan Heavy Industry Co., Ltd, Taiyuan 030024, China; 3.Shanxi Taizhong Shuzhi Technology Co., Ltd, Taiyuan 030024, China; 4. Shanxi Guohua Energy Co.,Ltd, Taiyuan 030000, China)

Received date: 2023-08-21

Revised date: 2023-10-18

Accepted date: 2023-10-18

Online published: 2023-08-21

Fold

摘要

针对游弋舰船或近水面航行的潜艇等目标在低帧率或视频图像中缺失部分帧情况下，跟踪目标帧与帧之间存在较大差距，导致跟踪精度下降、效率低的问题，提出一种基于YOLOv5与DeepSORT融合的水面航行多目标跟踪算法。首先，引入超分辨率重建网络对跟踪目标图像进行增强，以消除云雾及海浪对识别网络的干扰，使图像中目标特征清晰化；其次，在YOLOv5中引入ShuffleAttention注意力模块以增强识别网络对目标特征的提取能力；最后，在DeepSORT算法级联匹配中引入欧氏距离匹配替代IOU匹配，以此提升目标跟踪精度。仿真结果表明，本文算法的跟踪效果良好，改进的YOLOv5模型相对mAP50-95值提升了9.4%；在DeepSORT跟踪网络中，跟踪准确率对比优化前提升了8.11%。

关键词： 船舶; 水面航行; 多目标跟踪; 超分辨率重建; YOLOv5; DeepSORT; 欧氏距离匹配

本文引用格式

付诗文, 韩星程, 王黎明, 武国强, 王鸿儒, 马文, 王志勇 . 基于YOLOv5-DeepSORT融合的低帧率水面航行多目标跟踪算法[J]. 大连海事大学学报, 2024 , 50(1) : 94 -101 . DOI: 10.16411/j.cnki.issn1006-7736.2024.01.011

Abstract

A multi-target tracking algorithm for water navigation based on the fusion of YOLOv5 and DeepSORT was proposed to address the problem of significant differences between tracking target frames in low frame rates or missing frames in video images for targets such as cruising ships or submarines sailing near the water surface, resulting in decreased tracking accuracy and efficiency. Firstly, a super-resolution reconstruction network was introduced to enhance the tracking target image to eliminate the interference of clouds, fog, and waves on the recognition network and make the target features in the image clear. Secondly, the ShuffleAttention module was introduced in YOLOv5 to enhance the recognition network’s ability to extract target features. Finally, in the DeepSORT algorithm cascade matching, Euclidean distance matching was introduced instead of IOU matching to improve target tracking accuracy. Simulation results show that the tracking performance of the algorithm proposed is good, and the improved YOLOv5 model has increased the mAP50-95 value by 9.4%, and in the DeepSORT tracking network, the tracking accuracy has increased by 8.11% compared to before optimization.

Key words： ship; surface navigation; multi-target tracking; super resolution reconstruction; YOLOv5; DeepSORT; Euclidean distance matching

参考文献

[1]叶晨, 逯天洋, 肖潏灏，等. 海事监控视频舰船目标检测研究现状与展望[J].中国图象图形学报, 2022, 27(7): 2078-2093.
YE C, LU T Y, XIAO Y H, et al. Research status and prospect of maritime surveillance video ship target detection[J]. Journal of Image and Graphics, 2022, 27(7): 2078-2093.(in Chinese)
[2]MIHIR D, SIBILLA O, ADRIANO T, et al. Benchmarking YOLOv5 and YOLOv7 models with DeepSORT for droplet tracking applications[J]. The European Physical Journal E, 2023, 46:32 .
[3]SHAN Y X, LIU S H, ZHANG Y F, et al. LMD-TShip: vision based large-scale maritime ship tracking benchmark for autonomous navigation applications[J]. IEEE Access, 2021, 9: 74370-74384.
[4]ZHANG P C. Design and implementation of marine ship tracking system based on multi-target tracking algorithm[J]. Journal of Coastal Research, 2020, 110 (SI): 47-49.
[5]HU X J, ZHANG Q. Nighttime trajectory extraction framework for traffic investigations at intersections based on improved SSD and DeepSort[J]. Signal, Image and Video Processing, 2023,17: 2907-2914.
[6]CHEN Y T, WU B, LUO G Z, et al. Multi-target tracking algorithm based on YOLO+DeepSORT[J]. Journal of Physics: Conference Series, 2022, 2414: 012018.
[7]张铁栋, 李仁哲, 郎硕等. 基于DeepSORT的水下目标声学图像跟踪方法[J].华中科技大学学报(自然科学版), 2023(10): 50-56.
ZHANG T D, LI R Z, LANG S, et al. Acoustic image tracking method of underwater target based on DeepSORT[J].Journal of Huazhong University of Science and Technology (Natural Science Edition): 2023(10): 50-56.(in Chinese)
[8]于国莉, 桑金歌, 李俊荣. 基于改进卷积神经网络的舰船实时目标跟踪识别技术[J].舰船科学技术,2022, 44(21):152-155.
YU G L, SANG J G, LI J R. Real-time target tracking recognition technology for ships based on improved convolutional neural network[J].Ship Science and Technology, 2022, 44(21):152-155.(in Chinese)
[9]SONG H J, ZAHNG X H, SONG J, et al. Detection and tracking of safety helmet based on DeepSort and YOLOv5[J]. Multimedia Tools and Applications, 2022, 82: 10781-10794.
[10]洪宇, 叶瑞娟, 冯国富. 一种基于改进DeepSORT的淡水环境下鱼类计数方法[J/OL].(2023-07-03)[2023-07-10].海洋渔业, 2023. doi:10.13233/j.cnki.mar.fish.20230703.001.
HONG Y, YE R J, FENG G F. A fish counting method in freshwater environment based on improved DeepSORT[J/OL].(2023-07-03)[2023-07-10].Marine Fisheries, 2023. doi:10.13233/j.cnki.mar.fish.20230703.001. (in Chinese)
[11]任晨曦. 基于联合神经网络的水声目标识别技术研究[D].太原：中北大学, 2022.
REN C X. Research on underwater acoustic target recognition technology based on joint neural network[D]. Taiyuan: North University of China, 2022.(in Chinese)
[12]林迅, 姚力波, 孙炜玮, 等. 碎云环境下GF-4卫星对运动舰船的目标跟踪[J]. 航天返回与遥感, 2021, 42(5):127-139.
LIN X, YAO L B, SUN W W, et al. Target tracking of moving ships by GF-4 satellite under broken cloud environment[J]. Space Return & Remote Sensing, 2021,42(5):127-139. (in Chinese)
[13]GUO X T, ZUO M, YAN W J, et al. Behavior monitoring model of kitchen staff based on YOLOv5l and DeepSort techniques[J]. MATEC Web of Conferences, 2022. doi:10.1051/matecconf/202235503024.
[14]MOHAMMAD B S, ALI N, MAHMOUD S. Evaluating the spatial effects of environmental influencing factors on the frequency of urban crashes using the spatial Bayes method based on Euclidean distance and contiguity[J]. Transportation Engineering, 2023,12: 100181.

Options

文章导航

模态框（Modal）标题

摘要

本文引用格式

Abstract

参考文献