基于StrongSORT算法的羊只多目标跟踪方法[

doi:10.13733/j.jcam.issn.2095‑5553.2024.08.026

摘要/Abstract

摘要： 羊只的运动状态能够反映其健康状况，自动跟踪养殖场环境下的目标羊只是统计并分析其运动状态的前提。以圈养的羊只为试验对象，以YOLOv5-CBAM为前端检测器，结合目前比较先进的StrongSORT跟踪器，提出一种基于StrongSORT算法的羊只多目标跟踪方法。试验结果表明，在短视频跟踪中，对于10只羊的运动轨迹进行跟踪时，多目标跟踪准确度、多目标跟踪精确度、身份切换次数和IDF1值分别达到91.6%、0.269、52次和70.7%，与YOLOv5+StrongSORT算法相比，提出的YOLOv5-CBAM+StrongSORT算法的多目标跟踪准确度提高0.4%，多目标跟踪精确度基本不变，身份切换次数降低17.5%，IDF1提高3.2%；在长视频跟踪中，多目标跟踪准确度、多目标跟踪精确度、身份切换次数和IDF1值分别为57.3%、0.244、21次和47.9%，YOLOv5-CBAM+StrongSORT的优势主要体现在身份切换次数上，与YOLOv5+ByteTrack、YOLOv5+DeepSORT和YOLOv5+OCSORT相比，分别减少13次、10次和12次。

关键词: 羊只, 目标检测, 多目标跟踪, 改进YOLOv5, StrongSORT

Abstract: The behavior of sheep can reflect its health status and physiological stages. Automatic tracking of the objects for sheep in the farm environment is a prerequisite for statistics and analysis of its behavior. In this paper, captive sheep were used as the experimental subjects, and then a multiple object tracking method for sheep based on StrongSORT algorithm was proposed. It used YOLOv5-CBAM as the front⁃end detector, then combined the currently advanced StrongSORT tracker. The experimental results showed that, in the short video tracking, the multiple object tracking accuracy, multiple object tracking precision, the total number of identity switches and IDF1 of 10 sheep reached 91.6%, 0.269, 52 and 70.7%, respectively. Compared with the YOLOv5+StrongSORT algorithm, the multi‑object tracking accuracy of the YOLOv5-CBAM+StrongSORT algorithm proposed in this paper was improved by 0.4%, the multi‑object precision tracking was basically unchanged, the number of identity switching times was reduced by 17.5%, and the IDF1 value was increased by 3.2%. In the long video tracking, the above evaluation indicators were 57.3%, 0.244, 21 and 47.9%, respectively, and the advantages of YOLOv5-CBAM+StrongSORT were mainly reflected in the number of identity switches, which were reduced by 13, 10 and 12 times compared with YOLOv5+ByteTrack, YOLOv5+DeepSORT, and YOLOv5+OCSORT, respectively.

Key words: sheep, object detection, multiple object tracking, improved YOLOv5, StrongSORT

中图分类号:

赵晓霞, 程曼, 袁洪波. 基于StrongSORT算法的羊只多目标跟踪方法[[J]. 中国农机化学报, 2024, 45(8): 180-188.

Zhao Xiaoxia, Cheng Man, Yuan Hongbo. A multi‑object tracking method for sheep based on StrongSORT algorithm [J]. Journal of Chinese Agricultural Mechanization, 2024, 45(8): 180-188.

参考文献

[ 1 ] Walter T, Couzin I D. TRex, a fast multi‑animal tracking system with markerless identification, and 2D estimation of posture and visual fields [J]. Elife, 2021: 1-72.
[ 2 ] 金宇. 拥挤场景下的人体姿态估计与跟踪[D]. 天津: 河北工业大学, 2020.
Jin Yu. Human pose estimation and tracking in crowded scene [D]. Tianjin: Hebei University of Technology, 2020.
[ 3 ] 荣佩佩. 视频监控中运动目标的检测、跟踪与动作识别算法的研究[D]. 保定: 河北大学, 2016.
[ 4 ] 涂淑琴, 刘晓龙, 梁云, 等. 基于改进DeepSORT的群养生猪行为识别与跟踪方法[J]. 农业机械学报, 2022, 53（8）: 345-352.
Tu Shuqin, Liu Xiaolong, Liang Yun, et al. Behavior recognition and tracking method of group‑housed pigs based on improved DeepSORT algorithm [J]. Transactions of the Chinese Society for Agricultural Machinery, 2022, 53（8）: 345-352.
[ 5 ] Hassan-Vásquez J A, Maroto⁃Molina F, Guerrero⁃Ginel J E. GPS tracking to monitor the spatiotemporal dynamics of cattle behavior and their relationship with feces distribution [J]. Animals, 2022, 12（18）: 2383-2404.
[ 6 ] Barnes A L, Wickham S L, Admiraal R, et al. Characterization of inappetent sheep in a feedlot using radio⁃tracking technology [J]. Journal of Animal Science, 2018, 96（3）: 902-911.
[ 7 ] 骆焦煌, 宋长龙. 基于改进 CNN 算法的视觉图像目标跟踪研究[J]. 吉林大学学报（信息科学版）, 2023, 41（1）: 165-173.
Luo Jiaohuang， Song Changlong. Research on visual image target tracking based on improved convolution neural network algorithm [J]. Journal of Jilin University （Information Science Edition）, 2023, 41（1）: 165-173.
[ 8 ] 孟琭, 杨旭. 目标跟踪算法综述[J]. 自动化学报, 2019, 45（7）: 1244-1260.
Meng Lu, Yang Xu. A survey of object tracking algorithms [J]. Acta Automatica Sinica, 2019, 45（7）: 1244-1260.
[ 9 ] 黄智慧, 詹瑾, 赵慧民, 等. 基于深度学习的视觉目标跟踪算法浅析[J]. 广东技术师范学院学报, 2019, 40（3）: 28-36.
Huang Zhihui, Zhan Jin, Zhao Huimin, et al. Analysis of object tracking algorithms based on deep learning technologies [J]. Journal of Guangdong Polytechnic Normal Univercity, 2019, 40（3）: 28-36.
[10] Kashiha M, Bahr C, Ott S, et al. Automatic identification of marked pigs in a pen using image pattern recognition [J]. Computers and Electronics in Agriculture, 2013, 93: 111-120.
[11] 康熙, 张旭东, 刘刚, 等. 基于机器视觉的跛行奶牛牛蹄定位方法[J]. 农业机械学报, 2019, 50（S1）: 276-282.
Kang Xi, Zhang Xudong, Liu Gang, et al. Hoof location method of lame dairy cows based on machine vision [J]. Transactions of the Chinese Society for Agricultural Machinery, 2019, 50（S1）: 276-282.
[12] Bewley A, Ge Z, Ott L, et al. Simple online and realtime tracking [C]. USA: IEEE, 2016.
[13] Wojke N, Bewley A, Paulus D. Simple online and realtime tracking with a deep association metric [C]. USA: IEEE, 2017.
[14] Du Y, Zhao Z, Song Y, et al. StrongSORT: Make DeepSORT great again [J]. IEEE Transactions on Multimedia, 2023: 13514-13532.
[15] Emami P, Pardalos P M, Elefteriadou L, et al. Machine learning methods for data association in multi‑object tracking [J]. ACM Computing. Surveys, 2020, 53（4）: 1-33.
[16] 李琦, 尚绛岚, 李宝山. 基于 YOLOv3 和 DeepSORT 的草原牛跟踪系统[J]. 传感器与微系统, 2021, 40（6）: 83-85, 88.
Li Qi, Shang Jianglan, Li Baoshan. Grassland cattle tracking system based on YOLOv3 and DeepSORT [J]. Transducer and Microsystem Technologies, 2021, 40（6）: 83-85, 88.
[17] 张宏鸣, 汪润, 董佩杰, 等. 基于DeepSORT算法的肉牛多目标跟踪方法[J]. 农业机械学报, 2021, 52（4）: 248-256.
Zhang Hongming, Wang Run, Dong Peijie, et al. Beef cattle multi-target tracking based on DeepSORT algorithm [J]. Transactions of the Chinese Society for Agricultural Machinery, 2021, 52（4）: 248-256.
[18] Cheng M, Yuan H, Wang Q, et al. Application of deep learning in sheep behaviors recognition and influence analysis of training data characteristics training data characteristics on the recognition effect [J]. Computers and Electronics in Agriculture, 2022, 198: 107010.
[19] 杨国欣. 基于Adaboost算法和视觉显著性的羊只目标检测与计数方法研究[D]. 咸阳: 西北农林科技大学, 2019.

[1]	周一帆, 刘东洋, 周宇平. 基于改进FairMOT的实时群养生猪智能跟踪[J]. 中国农机化学报, 2024, 45(8): 49-59.
[2]	冯晓, 张辉, 刘运超, 张微, 李小红, 马中杰, . 基于改进YOLOv5s的硬核期葡萄簇检测[J]. 中国农机化学报, 2024, 45(8): 240-245.
[3]	周思捷, 刘天奇, 陈天华. 基于改进YOLOv5算法的水稻病害识别研究[J]. 中国农机化学报, 2024, 45(8): 246-253.
[4]	岳耀华, 张伟, 亓立强, . 基于改进YOLOv5的田间大豆花朵生长状态检测方法研究[J]. 中国农机化学报, 2024, 45(7): 188-193.
[5]	张延军, 赵建鑫, . 基于多路神经网络的黄花菜关节点检测算法研究[J]. 中国农机化学报, 2024, 45(7): 228-234.
[6]	张晨一, 张晓乾, 任振辉. 基于改进YOLOv3的自然场景中未成熟青苹果识别研究[J]. 中国农机化学报, 2024, 45(7): 243-248.
[7]	杨昊霖, 王其欢, 李华彪, 耿端阳, 武继达, 姚艳春. 基于改进YOLOv5的田间复杂环境障碍物检测[J]. 中国农机化学报, 2024, 45(6): 216-222.
[8]	薛霞, 刘鹏, 周文. 基于改进YOLOv8算法的实时细粒度植物病害检测[J]. 中国农机化学报, 2024, 45(5): 188-194.
[9]	罗鑫, 李加强, 何超. 基于改进YOLOv4的澳洲坚果视觉监测方法[J]. 中国农机化学报, 2024, 45(5): 217-222.
[10]	霍正瑞, 孙铁波. 基于改进YOLOv5算法的瓜蒌分级方法[J]. 中国农机化学报, 2024, 45(4): 100-107.
[11]	范才虎, 程曼, 袁洪波, 王媛, 蔡振江. 基于多角度Kinect v2的羊只三维模型重构方法研究[J]. 中国农机化学报, 2024, 45(3): 189-197.
[12]	黄威, 刘义亭, , 李佩娟, 陈光明, . 基于改进YOLOXS的苹果成熟度检测方法[J]. 中国农机化学报, 2024, 45(3): 226-232.
[13]	叶荣, 马自飞, 高泉, 李彤, 邵郭奇, 王白娟. 基于改进YOLOv5sECAASFF算法的茶叶病害目标检测[J]. 中国农机化学报, 2024, 45(1): 244-251.
[14]	王瑞彬, 杨世忠, 高升. 融合残差网络与注意力机制的草莓检测[J]. 中国农机化学报, 2024, 45(1): 266-273.
[15]	林森, , 许童羽, 葛禹豪, 马璟, 孙添龙, 赵春江. 基于改进YOLOv5l的设施番茄3D信息检测方法[J]. 中国农机化学报, 2024, 45(1): 274-284.