洗煤厂中工作人员的安全问题一直以来都被人们所重视，为此，洗煤厂中引入了智能化的视频监控，以便及时的发现监控视频中的异常状况，减少安全性问题的发生。但由于洗煤厂中的监控设备安装位置固定，大型设备较多，导致工作人员在日常工作中容易相互遮挡或被大型设备等遮挡，增加了人员识别与跟踪的难度。

针对遮挡导致的人员跟踪精度差、人员无法识别和错误识别的问题，结合了目标检测、目标跟踪和行人重识别技术，提出了一种基于YOLOv5s+DeepSORT+FastReID的多目标跨摄像机识别与跟踪的方法。具体的工作内容如下：

在人员跟踪部分使用DeepSORT目标跟踪算法作为主要的人员跟踪算法，选用YOLOv5s作为DeepSORT的检测器。首先针对YOLOv5s在进行人员检测时遇到的检测框重框的问题，采用EIOU-NMS(Efiicient Generalized Intersection Over Union Non Maximum Suppression)算法替换了原有的NMS(Non Maximum Suppression)算法，然后针对DeepSORT在人员跟踪时因遮挡导致的ID跳变的问题，将DeepSORT中原有的ReID(Re-identification)模型替换为FastReID的ReID模型，并在该模型中添加了注意力机制，进一步提升模型的特征提取能力。实验结果表明，改进后的YOLOv5s算法的准确率提升了0.8%，召回率提升了0.4%，平均精度均值提升了0.2%；添加了注意力机制的FastReID算法在Market1501数据集上的mAP提高了0.4%，Rank1提高了1.3%；改进后的DeepSORT算法的ID跳变次数减少了30%。

在人员识别部分选用FastReID行人重识别算法作为主要的人员识别算法，针对人员识别过程中因遮挡导致的人员无法识别和被错误识别的问题，采用添加注意力机制、建立动态人员图像库和一种运动估计的方法，实验结果表明结合三种方法之后的FastReID算法将人员识别过程中的None的次数与误识别次数均减少了70%。

经过实验验证，改进后的YOLOv5s+DeepSORT+FastReID算法能够实现洗煤厂场景下的人员跨摄像机识别与跟踪，在计算力高的设备中能够达到实时识别和跟踪的需求。

The safety of staff in coal washing plants has long been valued and for this reason, intelligent video surveillance has been introduced in coal washing plants to enable the timely detection of abnormal conditions in the surveillance video and to reduce the occurrence of safety problems. However, as the monitoring equipment in coal washing plants is installed in a fixed position, there are more large equipment devices, resulting in staff being easily obscured by each other or by large equipment, etc. in their daily work, increasing the difficulty of identifying and tracking personnel.

Aiming at the problems of poor person tracking accuracy, and unrecognisable and incorrect person identification due to occlusion, a multi-target cross-camera recognition and tracking method based on YOLOv5s+DeepSORT+FastReID is proposed by combining target detection, target tracking, and person re-identification techniques.The details of the work are as follows.

The DeepSORT target tracking algorithm was used as the main person tracking algorithm in the person tracking section, and YOLOv5s was selected as the DeepSORT detector. Firstly, the EIOU-NMS(Efiicient Generalized Intersection Over Union Non Maximum Suppression) algorithm was used to replace the original NMS(Non Maximum Suppression) algorithm for the problem of re-framing of detection frames encountered by YOLOv5s when performing person detection, and then the original ReID(Re-identification) model in DeepSORT was replaced with the ReID model of FastReID for the problem of ID jumping caused by occlusion during person tracking in DeepSORT, and an attention mechanism was added to the model to further improve the feature extraction capability. attention mechanism was added to the model to further improve the feature extraction capability of the model. The experimental results show that the improved YOLOv5s algorithm improves the accuracy by 0.8%, the recall by 0.4% and mAP by 0.2%; the FastReID algorithm with the added attention mechanism improves the mAP by 0.4% and the Rank1 1.3% on the Market1501 dataset; the improved DeepSORT algorithm reduces the number of ID jumps by 30%.

The FastReID pedestrian re-identification algorithm was chosen as the main person recognition algorithm in the person recognition section. The experimental results show that the FastReID algorithm reduces the number of None and false identifications by 70%.

After experimental verification, the improved YOLOv5s+DeepSORT+FastReID algorithm is able to achieve cross-camera identification and tracking of people in coal washing plant scenarios, and can achieve real-time identification and tracking requirements in devices with high computational power.

[1]袁红.数字视频监控技术在煤矿安全生产中的应用[J].能源与节能,2022(06):208-210.

[2]余京蕾.浅谈计算机视觉技术进展及其新兴应用[J].北京联合大学学报,2020,34(01):63-69.

[3]刘峰,郭林峰,赵路正.双碳背景下煤炭安全区间与绿色低碳技术路径[J].煤炭学报,2022,47(01):1-15.

[4]白路遥,闫浩良,吕林林,等.基于卫星定位的管道易水毁段预警技术[J].油气田地面工程,2021,40(12):1-4.

[5]李丽,郑嘉利,罗文聪,等.基于近端策略优化的RFID室内定位算法[J].计算机科学,2021,48(04):274-281.

[6]王行娟.基于WiFi的分级室内定位[J].电讯技术,2021,61(10):1291-1296.

[7]黄有山,候鸣,徐玲,等.基于ZigBee的室内定位方法分析和验证[J].智能物联技术,2018,1(02):23-27.

[8]蔡春昊,何元清,程德昊.基于蓝牙模块的机场室内定位技术研究与应用[J].无线互联科技,2020,17(20):169-170.

[9]张鹏,周代勇.基于UWB的洗煤厂定位方法研究[J].自动化与仪器仪表,2022(08):130-132+137.

[10]何惜琴,王食军,苏伟嘉.基于物联网的教学楼内人员检测系统的设计与实现[J].物联网技术,2022,12(09):35-38.

[11]张梅,荣昆,张啸.煤矿井下人员跟踪管理系统研究[J].煤矿机械,2020,41(12):172-175.

[12]Liu Xiaobai, Xu Yuanlu, Zhu Lei, et al. A Stochastic Attribute Grammar for Robust Cross-View Human Tracking[J]. IEEE Transactions on Circuits & Systems for Video Technology, 2018, 28(10):2884-2895.

[13]Ren Shaoqing, He Kaiming, Girshick R, et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6):1137-1149.

[14]从明芳，李子印，卢鸳，等.基于Mask R-CNN深度学习的羊绒羊毛纤维识别技术[J].现代纺织技术，2022,30(2):36-40,47.

[15]张建明,刘煊赫,吴宏林,等. 面向小目标检测结合特征金字塔网络的SSD改进模型[J]. 郑州大学学报（理学版）,2019,51(3):61-66,72.

[16]朱杰,辛月兰,孙可心. 二次特征融合的YOLO目标检测算法[J]. 计算机与数字工程,2021,49(5):914-919.

[17]Yang Tong, Zhang Xiangyu, Zhang Wenqiang, et al. MetaAnchor: Learning to Detect Objects with Customized Anchors[C]. Neural Information Processing Systems Montreal, 2018:318-328.

[18]Wang Jiaqi ,Chen Kai,Yang Shuo, et al. Region Proposal by Guided Anchoring[C]. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019:2965-2974.

[19]Law Hei,Deng Jia. Cornernet: Detecting objects as paired keypoints[J]. International Journal ofComputer Vision, 2018, 128(3): 642-656.

[20]Duan Kaiwen, Bai Song, Xie Lingxi, et al.CenterNet: Keypoint Triplets for Object Detection[C].2019 IEEE/CVF International Conference on Computer Vision(ICCV). 2019: 6568-6577.

[21]谭芳喜,肖世德,周亮君,等. 基于改进YOLOv3算法在道路目标检测中的应用[J]. 计算机技术与发展,2021,31(8):118-123.

[22]张长伦,张翠文,王恒友,等.基于注意力机制的NMS在目标检测中的研究[J].电子测量技术,2021,44(19):82-88.

[23]侯志强,刘晓义,余旺盛,等. 使用GIoU改进非极大值抑制的目标检测算法[J]. 电子学报,2021,49(4):696-705.

[24]Zhang Yifan,Ren Weiqiang, Zhang Zhang, et al. Focal and efficient IOU loss for accurate bounding box regression[J]. Computer Science, 2021, 12(1):1-10.

[25]Han Ruize, Zhao Jiewen, Feng Wei, et al. Complementary-View Co-Interest Person Detection[C].ACM Multimedia Conference 2020:2746-2754.

[26]Liu Caihong, Zhang Lei, Huang Hua. Visualization of Cross-View Multi-Object Tracking for Surveillance Videos in Crossroad[J]. Chinese Journal of Computers, 2018,41(1):221-235.

[27]彭建盛,许恒铭,李涛涛,等.生成式与判别式视觉目标跟踪算法综述[J].科学技术与工程,2021,21(35):14871-14881.

[28]Li Bo, Yan Junjie, Wu Wei, et al. High performance visual tracking with siamese region proposal network[C].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 8971-8980.

[29]Nicolai Wojke, Alex Bewley and Dietrich Paulus, "Simple online and realtime tracking with a deep association metric"[C]. 2017 IEEE International Conference on Image Processing (ICIP), 2017:3645-3649.

[30]胡晓强,魏丹,王子阳,等.基于时空关注区域的视频行人重识别[J].计算机工程,2021,47(6):277-283.

[31]Wu Ancong , Zheng Weishi , Gong Sshaogang , et al. RGB-IR Person Re-identification by Cross-Modality Similarity Preservation[J]. International Journal of Computer Vision, 2020, 128(8):1765-1785.

[32]王瀛,徐奔,左方.基于HOF-CNN和HOG特征的视频动作识别系统[J].计算机仿真,2022,39(06):179-182+318.

[33]Liu Donghaisheng, Han Shudong, Chen Yang, et al. Foreground-guided textural-focused person re-identification[J]. Neurocomputing, 2022,8(28):235-248.

[34]Xuan Zhang, Hao Luo, Xing Fan, et al. Aligned ReID: Surpassing human-level performance in person re-identification[J].Computer Vision and Pattern Recognition,2017,36(2):1383-1387.

[35]He Lingxiao, Liao Xingyu, Liu Wu,et al.FastReID: A Pytorch Toolbox for General Instance Re-identification[J].Computer Vision and Pattern Recognition, 2020,14(3):407-414.

[36]赵安新,杨金桥,史新国,等.视觉行人重识别研究方法分析及评价指标探讨[J].西安科技大学报,2022,42(05):1003-1012.

[37]熊炜,乐玲,周蕾,等.基于多层级特征融合的行人重识别算法[J].光电子·激光,2021,32(8):872-878.

[38]Wang Linlin,Wang Wei,Liang Fengmei.Pedestrianre-identincation based on fbature fusion of overlappedstripes[J].Modern Electronics Technique, 2019,41(10):175-178.

[39]徐同文,白宗文,杨延宁,等.基于光照不变性颜色特征的行人再识别方法[J].电子设计工程,2021,29(14):154-158.

[40]Tang Hongzhong，Chen Tianyu，Deng Shijun, et al. Multi-Level Discriminative Dictionary Learning Method for Cross-View Person Re-Identification[J].Journal of Computer-Aided Design & Computer Graphics, 2020, 32(9):1430-1441.

[41]Hermans Alexander, Beyer Lucas, Leibe Bastian. In Defense of the Triplet Loss for Person Re-Identification[C]. Computer Vision and Pattern Recognition, 2017:1703-1719.

[42]殷雨昌,王洪元,陈莉,等. 基于单标注样本的多损失学习与联合度量视频行人重识别[J]. 计算机应用,2022,42(3):764-769. 

[43]Sun Yifan, Cheng Changmao, Zhang Yuhan, et al."Circle Loss: A Unified Perspective of Pair Similarity Optimization"[C]. Computer Vision and Pattern Recognition, 2020: 6397-6406.

[44]刘晓蓉,李小霞,秦昌辉. 融合多尺度对比池化特征的行人重识别方法[J]. 计算机工程,2022,48(4):292-298.

[45]薛丽霞,朱正发,汪荣贵,等.基于多分区注意力的行人重识别方法[J].光电工程,2020,47(11):23-32.

[46]宋晓茹,杨佳,高嵩,等.基于注意力机制与多尺度特征融合的行人重识别方法[J].科学技术与工程，2022,22(04):1526-1533.

[47]Zheng Zhaohui , Wang Ping , Ren Dongwei , et al. Enhancing Geometric Factors in Model Learning and Inference for Object Detection and Instance Segmentation[J]. 2020,52 (8) :8574-8586.

[48]Li Jingbei, Meng Yi ,Wu Zhiyong, et al. "NEUFA: Neural Network Based End-to-End Forced Alignment with Bidirectional Attention Mechanism"[C].International Conference on Acoustics, Speech and Signal Processing, 2022:8007-8011.

[49]张海燕,张富凯,袁冠等.多姿态图像生成的行人重识别算法研究[J].计算机工程与应用:2022,19(3):1-12.

[50]Gustavo Mercado.The Filmmaker's Eye: Learning (and Breaking) the Rules of Cinematic Composition[M].New York:Routledge,2013.