作者：李琦, 尚绛岚, 李宝山

作者单位：内蒙古科技大学信息工程学院，内蒙古 包头 014010

关键词：目标检测;YOLOv3;目标跟踪;Deep SORT;羊群计数

摘要：

为解决当前国内牧场羊群数量由人工统计完成导致的人工成本高、统计效率低的问题，实验采用YOLOv3目标检测算法与Deep SORT跟踪算法相结合，基于双线计数法实现草原羊的自动计数。结果表明：针对标定的羊群头部数据集，在原始YOLOv3检测算法的基础上，采用K-means聚类方法进行聚类分析，改进YOLOv3检测算法的初始候选框，在测试集上检测准确度为90.12%，较原始YOLOv3提高8.57%；利用YOLOv3+Deep SORT的跟踪方法对草原羊头部目标进行跟踪，与Deep SORT跟踪算法的结果进行对比，跟踪成功率提高11.77%，中心点误差降低1.43%。实验在内蒙古苏尼特左旗合作牧场对草原羊进行计数并与真实值比较，计数精度较高，满足实验要求。说明基于头部图像特征的草原羊自动计数方法可以作为一种解决方案进行推广应用。

Method for grassland sheep automatic counting based on head image features
LI Qi, SHANG Jianglan, LI Baoshan
School of Information Engineering, Inner Mongolia University of Science and Technology, Baotou 014010, China
Abstract: In order to solve the problems of high labor cost and low statistical efficiency caused by the current completion of domestic pasture sheep population by manual statistics, the experiment uses the combination of YOLOv3 target detection algorithm and Deep SORT tracking algorithm to realize the automatic counting of prairie sheep based on the two-line counting method. The experimental results show that for the constructed sheep head dataset, K-means clustering method is used to perform cluster analysis based on the original YOLOv3 detection algorithm to improve the initial candidate box. The detection accuracy on the test set is 90.12%, which is 8.57% higher than the original YOLOv3. YOLOv3+Deep SORT algorithm is used to track the head of the grassland sheep. Compared with the results of the Deep SORT, the tracking success rate is increased by 11.77%, and the center point error is reduced by 1.43%. The experiment counted the grassland sheep in Sunite Zuoqi cooperative ranch in Inner Mongolia and compared it with the true value. The counting accuracy is higher, which meets the experimental requirements. It shows that the automatic counting method of grassland sheep based on head image features can be promoted and applied as a solution.
Keywords: target detection;YOLOv3;target tracking;Deep SORT;sheep counting
2020, 46(11):20-24  收稿日期: 2020-04-03;收到修改稿日期: 2020-05-29
基金项目: 内蒙古自然科学基金（2019MS06021）；内蒙古自治区科技成果转化项目（CGZH2018041）
作者简介: 李琦(1973-),男,陕西榆林市人,教授,研究方向为智能优化控制和工业远程控制
参考文献
[1] 张雅俊, 高陈强, 李佩, 等. 基于卷积神经网络的人流量统计[J]. 重庆邮电大学学报(自然科学版), 2017, 29(2): 265-271
[2] 时增林, 叶阳东, 吴云鹏, 等. 基于序的空间金字塔池化网络的人群计数方法[J]. 自动化学报, 2016, 42(6): 866-874
[3] 王小鹏, 渠燕红, 马鹏, 等. 基于头部图像特征的人流计数方法[J]. 铁道学报, 2019, 41(2): 78-83
[4] 张震, 李进钢. 基于OpenCV的公交客流计数方法[J]. 现代电子技术, 2017, 40(2): 16-20
[5] 马永杰, 宋晓凤. 基于YOLO和嵌入式系统的车流量检测[J]. 液晶与显示, 2019, 34(6): 613-618
[6] 徐娇, 赵其杰, 张君绍,等. 基于多特征匹配跟踪的深海发光浮游生物自动计数方法[J]. 海洋科学, 2019, 43(5): 64-70
[7] 郭俊, 郑堤, 陈俊华, 等. 基于机器视觉跟踪的计数算法[J]. 传感器与微系统, 2018, 37(2): 154-157+160
[8] 田磊. 羊群个数检测系统设计[D]. 呼和浩特: 内蒙古大学, 2019.
[9] REDMON J, FARHADI A. YOLOv3: An incremental improvement[C]//IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, IEEE, 2018: 89-95.
[10] WOJKE  N, BEWLEY  A, PAULUS  D. Simple online and realtime tracking with a deep association metric[C]//IEEE International Conference on Image Processing. IEEE, 2017: 3645-3649.
[11] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.