煤炭是我国工业发展的重要能源，每年都有着非常大的开采规模。但由于煤矿井下亮度低、粉尘含量高、作业环境恶劣，事故时有发生。因此，矿井工作人员的安全保障尤为重要，煤矿井下行人检测与跟踪对煤矿的安全生产具有重要意义。本文对矿井下监控装置采集到的视频图像进行手动标注，构建了小型的煤矿井下行人检测与跟踪数据集，并在此基础上对煤矿井下行人检测与跟踪问题进行研究，主要内容如下：

（1）针对煤矿井下行人检测精度不足、实时性要求高、环境条件差、行人状态复杂等问题，提出一种改进的FCOS煤矿井下行人检测算法。该模型使用轻量级卷积神经网络ShuffleNet V2替换FCOS检测算法中的骨干网络ResNet-50，将原始网络中的特征金字塔结构改进为自上而下和自下而上的路径聚合网络，同时利用由两组深度可分离卷积组成的轻量化检测头替换原始FCOS网络的检测头。在实验训练过程中，通过对井下行人检测数据进行尺度和颜色数据增强来提升模型的泛化能力与鲁棒性。实验结果显示，改进的FCOS可以更好地实现精度与速度之间的平衡，该算法在基本不损失精度的情况下，mAP达51.9%，FPS可以达到100帧/s。

（2）基于改进的FCOS目标检测器的输出结果，使用卡尔曼滤波器进行预测与更新，并利用改进的匈牙利匹配算法对跟踪轨迹与检测结果进行数据关联与匹配。具体来说，为了充分利用因遮挡等问题导致检测器所输出的一部分置信度较低的检测框，将检测器的输出结果通过阈值设定的方式，分为高分检测框集合和低分检测框集合。首先将高分检测框集合与原跟踪轨迹进行关联和匹配，然后将未能成功与高分检测框匹配的跟踪轨迹与低分检测框集合进行关联与匹配，挖掘出低分检测框中有价值的目标信息，从而降低漏检并提高跟踪轨迹的连续性。通过在自构建的多个场景的煤矿井下数据集上进行实验，结果显示，改进的跟踪算法能够显著的提升跟踪器的跟踪性能。

Coal is an important energy source for China's industrial development and has a very large mining scale every year. However, due to low brightness, high dust content and bad working environment, accidents often occur in coal mines. Therefore, the safety guarantee of mine workers is particularly important, and the detection and tracking of underground pedestrians is of great significance to the safety production of coal mines. In this paper, the video images collected by the underground monitoring device are manually annotated, and a small data set of underground coal mine pedestrian detection and tracking is constructed. Based on this, the detection and tracking of underground coal mine pedestrian is studied. The main contents are as follows:

(1) An improved FCOS pedestrian detection algorithm is proposed to solve the problems of insufficient detection accuracy, high real-time requirement, poor environmental conditions and complex pedestrian status in underground coal mine. In this model, a lightweight convolutional neural network ShuffleNet V2 is used to replace the backbone network ResNET-50 in FCOS detection algorithm, and the feature pyramid structure in the original network is improved into a top-down and bottom-up path aggregation network. At the same time, the detection head of the original FCOS network is replaced by a lightweight detection head composed of two sets of depth-separable convolution. In the course of experimental training, the generalization ability and robustness of the model are improved by enhancing the scale and color data of downhole pedestrian detection data. Experimental results show that the improved FCOS algorithm can achieve a better balance between accuracy and speed. The mAP of the algorithm can reach 51.9% and the FPS can reach 100 frames /s without losing accuracy.

(2) Based on the output results of the improved FCOS target detector, Kalman filter is used for prediction and update. The improved Hungarian matching algorithm is used to correlate and match the tracking trajectories and detection results. Specifically, in order to make full use of some detection frames with low confidence output by the detector due to occlusion and other problems, the output results of the detector are divided into high-score detection frame set and low-score detection frame set through threshold setting. Firstly, the high-score detection frame set is associated and matched with the original tracking track, and then the tracking track of the high-score detection frame that has not been matched successfully before is associated and matched with the low-score detection frame set, mining the valuable target information in the low-score detection frame, so as to reduce the missed detection and improve the continuity of tracking track. Experimental results show that the improved tracking algorithm can significantly improve the tracking performance of the tracker.

[1] 董茜茜. 矿井环境监测中无线传感网分簇路由协议研究[D]. 南京: 南京邮电大学, 2014.

[2] Åstrand M, Jakobsson E, Lindfors M, et al. A system for underground road condition monitoring[J]. International Journal of Mining Science and Technology, 2020, 30(3): 405-411.

[3] Lal N, Kumar S, Chaurasiya V K. A road monitoring approach with real-time capturing of events for efficient vehicles safety in smart city[J]. Wireless Personal Communications, 2020, 114(3): 657-674.

[4] 刘文江. 矿井人员井下位置跟踪监控系统的实现方案比较研究[J]. 科技创新与应用, 2012 (08Z): 29-30.

[5] 李晓建. 矿井人员目标检测与跟踪算法的研究与实现[D]. 山东: 山东科技大学, 2020.

[6] 刘贝. 矿井人员目标检测与跟踪研究[D]. 西安: 西安科技大学, 2020.

[7] 耿蒲龙, 宋建成, 刘旭飞, 等. 基于RGB颜色空间的矿井运动目标检测及跟踪方法[J]. 太原理工大学学报, 2017, 48(6): 963-968.

[8] 杨铮. 基于视频矿井下的人员计数算法研究[D]. 武汉: 武汉理工大学, 2013.

[9] 李首滨. 煤炭工业互联网及其关键技术[J]. 煤炭科学技术, 2020, 48(07): 98-108.

[10] 张帆, 李闯. 面向智能矿山与新工科的数字孪生技术研究[J]. 工矿自动化, 2020, 46(5): 15-20.

[11] 胡青松, 杨维, 丁恩杰, 等. 煤矿应急救援通信技术的现状与趋势[J]. 通信学报, 2019, 40(05): 163-179.

[12] 牛化康, 何小海, 汪晓飞, 等. 一种改进的ViBe目标检测算法[J]. 四川大学学报 (工程科学版), 2014, 46(S2): 104-108.

[13] Gang L, Shangkun N, Yugan Y, et al. An improved moving objects detection algorithm[C]//International Conference on Wavelet Analysis and Pattern Recognition. Los Alamitos: IEEE, 2013: 96-102.

[14] Lu X, Xu C, Wang L, et al. Improved background subtraction method for detecting moving objects based on GMM[J]. IEEJ Transactions on Electrical and Electronic Engineering, 2018, 13(11): 1540-1550.

[15] Wan W, Tang S, Zhang H. Moving object detection based on high-speed video sequence images[C]//Proceedings of the IEEE Joint International Information Technology and Artificial Intelligence Conference. Los Alamitos: IEEE 2019: 906-910.

[16] Medvedeva E. Moving object detection in noisy images[C]//Proceedings of the Mediterranean Conference on Embedded Computing. Los Alamitos: IEEE, 2019: 1-4.

[17] Redmon J, Divvala S, Girshick R, et al. You only look once: Unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos: IEEE, 2016: 779-788.

[18] Liu W, Anguelov D, Erhan D, et al. Ssd: Single shot multibox detector[C]//Proceedings of the European Conference on Computer Vision. Cham: Springer, 2016: 21-37.

[19] Ren S, He K, Girshick R, et al. Faster r-cnn: Towards real-time object detection with region proposal networks[J]. Advances in Neural Information Processing Systems, 2015, 91-99.

[20] Xia S, Peng D, Meng D, et al. A fast adaptive k-means with no bounds[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020.

[21] Duan K, Bai S. Centernet: Keypoint triplets for object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Los Alamitos: IEEE, 2019: 6569-6578.

[22] Law H, Deng J. Cornernet: Detecting objects as paired keypoints[C]//Proceedings of the European Conference on Computer Vision. Cham: Springer, 2018: 734-750.

[23] Tian Z, Shen C. Fcos: Fully convolutional one-stage object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Los Alamitos: IEEE, 2019: 9627-9636.

[24] Huang C, Wu B, Nevatia R. Robust object tracking by hierarchical association of detection responses[C]//Proceedings of the European Conference on Computer Vision. Berlin, Heidelberg: Springer, 2008: 788-801.

[25] Milan A, Roth S, Schindler K. Continuous energy minimization for multitarget tracking[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 36(1): 58-72.

[26] Aguilar W G, Luna M A, Moya J F, et al. Pedestrian detection for UAVs using cascade classifiers with meanshift[C]//Proceedings of the IEEE International Conference on Semantic Computing. Los Alamitos: IEEE, 2017: 509-514.

[27] Wang Z, Yang X, Xu Y, et al. CamShift guided particle filter for visual tracking[J]. Pattern Recognition Letters, 2009, 30(4): 407-413.

[28] 刘美枝, 杨磊, 高海. 结合角点特征的CamShift 目标跟踪算法研究[J]. 山西大同大学学报: 自然科学版, 2019, 35(5): 14-18.

[29] 翟卫欣, 程承旗. 基于Kalman滤波的Camshift运动跟踪算法[J]. 北京大学学报: 自然科学版, 2015(5): 799-804.

[30] Chen K, Song X, Zhai X, et al. An integrated deep learning framework for occluded pedestrian tracking[J]. IEEE Access, 2019(7): 26060-26072.

[31] Truong M T N, Kim S. A tracking-by-detection system for pedestrian tracking using deep learning technique and color information[J]. Journal of Information Processing Systems, 2019, 15(4): 1017-1028.

[32] 董观利, 宋春林. 基于视频的矿井行人越界检测系统[J]. 工矿自动化, 2017(2): 29-34.

[33] 原磊明, 王海斌, 刘旭飞,等. 矿井人身安全视频防护系统设计[J]. 煤炭技术, 2018, 37(11):3.

[34] 邹斐. 煤矿井下运动目标检测与跟踪研究[D]. 西安: 西安科技大学, 2018.

[35] Bau D, Zhu J Y, Strobelt H, et al. Understanding the role of individual units in a deep neural network[J]. Proceedings of the National Academy of Sciences, 2020, 117(48): 30071-30078.

[36] Kamilaris A, Prenafeta-Boldú F X. Deep learning in agriculture: A survey[J]. Computers and Electronics in Agriculture, 2018, 147(5): 70-90.

[37] Zhang C, Chen T. A survey on image-based rendering-representation, sampling and compression[J]. Signal Processing: Image Communication, 2004, 19(1): 1-28.

[38] Wang Z, Yang Z. Review on image-stitching techniques[J]. Multimedia Systems, 2020, 26(4): 413-430.

[39] Iliev A, Kyurkchiev N, Markov S. On the Approximation of the step function by some sigmoid functions[J]. Mathematics and Computers in Simulation, 2017, 133(3): 223-234.

[40] Fan, Engui. Extended tanh-function method and its applications to nonlinear equations[J]. Physics Letters A, 2000, 277(4): 212-218.

[41] Banerjee C, Mukherjee T, Pasiliao E. Feature representations using the reflected rectified linear unit (RReLU) activation[J]. Big Data Mining and Analytics, 2020, 3(2): 102-120.

[42] Jiang X, Pang Y, Li X, et al. Deep neural networks with elastic rectified linear units for object recognition[J]. Neurocomputing, 2018, 275(3): 1132-1139.

[43] 孙继平. 煤矿安全监控系统联网技术研究[J]. 煤炭学报, 2009, 34(11): 1546-1549.

[44] 李伟山, 卫晨, 王琳. 改进的 Faster RCNN 煤矿井下行人检测算法[J]. 计算机工程与应用, 2019, 55(4): 200-207.

[45] Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos: IEEE, 2005: 886-893.

[46] Zhang Y, Gao J, Zhou H. Breeds classification with deep convolutional neural network[C]//Proceedings of the 2020 12th International Conference on Machine Learning and Computing. 2020: 145-151.

[47] Dai J, Li Y, He K, et al. R-fcn: Object detection via region-based fully convolutional networks[C]//Proceedings of the Advances in Neural Information Processing Systems. California: NIPS, 2016: 379-387.

[48] Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE International Conference on Computer Vision. Los Alamitos: IEEE, 2017: 2980-2988.

[49] K. He, X. Zhang, S. Ren and J. Sun, Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos: IEEE, 2016:770-778.

[50] T. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan and S. Belongie, Feature pyramid networks for object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos: IEEE, 2017:936-944.

[51] S. Liu, L. Qi, H. Qin, J. Shi and J. Jia, Path aggregation network for instance segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos: IEEE, 2018: 8759-8768.

[52] Ma N, Zhang X, Zheng H T, et al. Shufflenet v2: Practical guidelines for efficient cnn architecture design[C]//Proceedings of the European Conference on Computer Vision. Berlin. Heidelberg: Springer, 2018: 116-131.

[53] 王娇娇, 刘政怡, 李辉. 特征融合与objectness加强的显著目标检测[J]. 计算机工程与应用, 2017, 53(2): 195-200.

[54] A Welch G F. Kalman filter[J]. Computer Vision: A Reference Guide, 2020: 1-3.

[55] Thanh L B. Data Association for Multi-Object Tracking Using Assignment Algorithms[C]//2021 International Conference Engineering and Telecommunication. IEEE, 2021: 1-5.