日常生活中自动扶梯是运送乘客十分常见的设施，在商场、地铁、机场、医院等公共场所被广泛使用。乘客摔倒事故已成为自动扶梯伤人事件的主要原因，传统自动扶梯日常管理消耗人工较多，当遇到突发状况难以立即被发现，常常因为无法及时按下“紧急停止按钮”终止扶梯运行，从而造成连续翻滚等重大人身伤害，因此实现自动扶梯智能化监控管理势在必行。

（1）具有坡度的自动扶梯运行环境更复杂，行人较多，局部遮挡情况频发，视频采集角度不断变化，传统的人体姿态特征摔倒检测算法效果不佳，检测速度较慢。因此融合Swin Transformer和YOLOX深度学习算法的优秀特性，本文提出了一种基于SwinT-YOLOX网络模型的自动扶梯行人摔倒检测算法。改进算法采用Swin Transformer模型作为骨干网络，颈部网络使用融合CBAM注意力机制的YOLOX模型，进一步提升模型特征图的多样性和表达能力。此外，利用FReLU视觉激活函数改进网络模块，从而获得更优秀的特征检测性能。本文模拟自动扶梯行人摔倒事件，在商场、地铁、医院和机场等场所共采样300段视频序列构建数据集。针对自建扶梯行人摔倒数据集，实验结果表明本文改进的SwinT-YOLOX自动扶梯摔倒检测算法能够快速、精准的检测到乘客摔倒事故发生，平均检测精度达到95.92%，相较于原始YOLOX算法提升了3.26%，并且可以实现实时检测，检测速率达到24fps左右。

（2）为了实现改进算法能有效部署在嵌入式硬件平台上，本文采用TensorRT优化器在NVIDIA Jetson TX2嵌入式平台进行推理部署优化，同时使用QT5开发扶梯监控管理软件界面。设计搭建的自动扶梯行人摔倒智能监控系统能够实时检测自动扶梯中的摔倒行为，并能针对异常行为发出语音警报和控制扶梯安全应急措施以保证乘客安全。

本文实现自动扶梯行人摔倒检测算法能够快速、精准的检测到乘客摔倒事故发生，监控管理平台可以及时发出预警信息，并立即实施紧急停车命令，确保乘客安全。扶梯智能监控系统可以全天高效检测，显著减轻扶梯日常安全管理人员的工作。

Escalator is a very common facility for transporting passengers in daily life. It is widely used in shopping malls, subways, airports, hospitals and other public places. Passenger fall accidents have become the main cause of escalator injuries. The daily management of traditional escalators consumes a lot of manpower. It is difficult to be found immediately when an emergency occurs. It is often impossible to press the ' emergency stop button ' in time to terminate the escalator operation, resulting in continuous rolling and other major personal injuries. Therefore, it is imperative to realize intelligent monitoring and management of escalators.

(1) The escalator with slope has more complex operating environment, more pedestrians, frequent partial occlusion, and constantly changing video acquisition angles. The traditional human posture feature fall detection algorithm is not effective and the detection speed is slow. Therefore, combining the excellent characteristics of Swin Transformer and YOLOX deep learning algorithm, this paper proposes an escalator pedestrian fall detection algorithm based on SwinT-YOLOX network model. The improved algorithm uses the Swin Transformer model as the backbone network, and the neck network uses the YOLOX model that integrates the CBAM attention mechanism to further improve the diversity and expression ability of the model feature map. In addition, the FReLU visual activation function is used to improve the network module to obtain better feature detection performance. In this paper, the escalator pedestrian fall event is simulated, and 300 video sequences are sampled in shopping malls, subways, hospitals and airports to construct data sets. Aiming at the self-built escalator pedestrian fall data set, the experimental results show that the improved SwinT-YOLOX escalator fall detection algorithm in this paper can quickly and accurately detect the occurrence of passenger fall accidents. The average detection accuracy reaches 95.92 %, which is 3.26 % higher than the original YOLOX algorithm, and can achieve real-time detection. The detection rate reaches about 24 fps.

(2) In order to realize the effective deployment of the improved algorithm on the embedded hardware platform, this paper uses the TensorRT optimizer to optimize the reasoning deployment on the NVIDIA Jetson TX2 embedded platform, and uses QT5 to develop the escalator monitoring management software interface. The designed escalator pedestrian fall intelligent monitoring system can detect the fall behavior in the escalator in real time, and can issue voice alarms for abnormal behaviors and control escalator safety emergency measures to ensure passenger safety.

In this paper, the escalator pedestrian fall detection algorithm can quickly and accurately detect the occurrence of passenger fall accidents. The monitoring and management platform can issue early warning information in time and immediately implement emergency parking orders to ensure passenger safety. The escalator intelligent monitoring system can detect efficiently throughout the day and significantly reduce the work of escalator daily safety management personnel.

[1]马爱萍.自动扶梯事故频发原因分析及对策探讨[J].科技信息,2013(25): 258-259.

[2]黄凯奇,陈晓棠,康运锋,等.智能视频监控技术综述[J].计算机学报,2015, 38(6):1093-1118.

[3]王桔,张斌,应征,等.基于机器视觉的自动扶梯梯级测速方法[J].中国测试, 2019,45(8):44-49.

[4]范进泉.谈智能视频监控系统对提高自动扶梯乘坐安全的重要性[J].中国电梯,2022,33(12):40-43.

[5]Gutiérrez J, Rodríguez V, Martin S. Comprehensive review of vision-based fall detection systems[J]. Sensors, 2021, 21(3), 947:1-50.

[6]Yu-qing C, Zhan-zhuang H E, Zhong M A, et al. Intelligent target detection algorithm for embedded FPGA[J]. 微电子学与计算机, 2021, 38(6): 87-92.

[7]Raghunandan A, Raghav P, Aradhya H V R. Object detection algorithms for video surveillance applications [C]//2018 International Conference on Communication and Signal Processing (ICCSP). IEEE, 2018: 0563-0568.

[8]Viola P, Jones M. Rapid object detection using a boosted cascade of simple features[C]//Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001. 1: I-I.

[9]Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]// 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR).1: 886-893.

[10]Pisner D A, Schnyer D M. Support vector machine[M]//Machine Learning. Academic Press, 2020: 101-121.

[11]Felzenszwalb P, McAllester D, Ramanan D. A discriminatively trained, multiscale, deformable part model[C]//2008 IEEE Conference on Computer Vision and Pattern Recognition. 2008: 1-8.

[12]Yuan Z W, Zhang J. Feature extraction and image retrieval based on AlexNet [C]//Eighth International Conference on Digital Image Processing (ICDIP 2016). SPIE, 2016, 10033: 65-69.

[13]Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2014: 580-587.

[14]Girshick R. Fast r-cnn[C]//Proceedings of the IEEE International Conference on Computer Vision. 2015: 1440-1448.

[15]He K, Zhang X, Ren S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9): 1904-1916.

[16]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, 28.

[17]Dai J, Li Y, He K, et al. R-fcn: Object detection via region-based fully convolutional networks[J]. Advances in Neural Information Processing Systems, 2016, 29.

[18]Li Z, Peng C, Yu G, et al. Light-head r-cnn: In defense of two-stage object detector[J]. arXiv preprint arXiv:1711.07264, 2017.

[19]Lin T Y, Dollár P, Girshick R, et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017: 2117-2125.

[20]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. 2016: 779-788.

[21]Liu W, Anguelov D, Erhan D, et al. Ssd: Single shot multibox detector[C]// Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14. Springer International Publishing, 2016: 21-37.

[22]Redmon J and Farhadi A. Yolo9000: better, faster, stronger[C]. Proceedings of Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017:7263–7271.

[23]Redmon J, Farhadi A. Yolov3: An incremental improvement[J]. arXiv preprint arXiv:1804.02767, 2018.

[24]Bochkovskiy A, Wang C Y, Liao H Y M. Yolov4: Optimal speed and accuracy of object detection[J]. arXiv preprint arXiv:2004.10934, 2020.

[25]Zhu X, Lyu S, Wang X, et al. TPH-YOLOv5: Improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenarios[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021: 2778-2788.

[26]Ge Z, Liu S, Wang F, et al. YOLOX: Exceeding YOLO Series in 2021[J]. arXiv preprint arXiv:2107.08430,2021.

[27]Li C, Li L, Jiang H, et al. YOLOv6: A single-stage object detection framework for industrial applications[J]. arXiv preprint arXiv:2209.02976, 2022.

[28]Wang C Y, Bochkovskiy A, Liao H Y M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[J]. arXiv preprint arXiv:2207.02696, 2022.

[29]Mirmahboub B, Samavi S, Karimi N, et al. Automatic monocular system for human fall detection based on variations in silhouette area[J]. IEEE Transactions on Biomedical Engineering, 2012, 60(2): 427-436.

[30]Ma X, Wang H, Xue B, et al. Depth-based human fall detection via shape features and improved extreme learning machine[J]. IEEE Journal of Biomedical and Health Informatics, 2014, 18(6): 1915-1922.

[31]Xu JY, Lian JX. Fall Behavior Detection Method Based on Human Behavior Model[J]. Computer Systems and Applications, 2020, 29(6): 189-195.

[32]Chen W M, Jiang Z J, Guo H L, et al. Fall detection based on key points of human-skeleton using openpose [J]. Symmetry, 2020, 12(5), 744:1-17

[33]马子越,彭瑞阳,孙晓晗,王钰泽,李欣悦,孔祥勇.基于OpenPose的人体姿态估计技术研究综述[J].软件导刊, 2022, 21(11): 247-252.

[34]卫少洁,周永霞. 一种结合Alphapose和LSTM的人体摔倒检测模型[J].小型微型计算机系统, 2019,40(9): 1886-1890.

[35]马敬奇,雷欢,陈敏翼.基于AlphaPose优化模型的老人跌倒行为检测算法[J].计算机应用,2022,42(1): 294-301.

[36]A.ARaza, M.H. Yousaf,S.A.Velastin. Human Fall Detection using YOLO: A Real-Time and AI-on-the-Edge Perspective[C]//12th International Conference on Pattern Recognition Systems (ICPRS), France: IEEE Pres, 2022: 1-6.

[37]Y. Yin, L. Lei, M. Liang, X. Li, et al. Research on Fall Detection Algorithm for the Elderly Living Alone Based on YOLO[C]//IEEE International Conference on Emergency Science and Information Technology (ICESIT 2021), CHINA: IEEE Pres, 2021: 403-408.

[38]王晓雯, 梁博, 刘芳芳. 基于注意力机制与加权盒函数的YOLOv5的行人摔倒检测算法[J].山西大学学报(自然科学版), 2023: 1-9.

[39]Zhao X. Research on the application of OpenPose in escalator safety systems[C]// 5th International Conference on Advanced Algorithms and Control Engineering (ICAACE 2022). CHINA: IOP, 2022,2258(1),012053:1-8

[40]Liu S, An Z, Wang N, et al. Research on elevator passenger fall detection based on machine vision[C]//IOP Conference Series: Earth and Environmental Science. CHINA: IOP, 2021, 791(1), 012108: 1-7.

[41]Jiao Z, Lei H, Zong H, et al. Potential escalator-related injury identification and prevention based on multi- module integrated system for public health[J]. Machine Vision and Applications, 2022, 33(2): 1-12.

[42]滕安. 基于人体姿态识别的行人乘坐自动扶梯跌倒检测方法的研究[D]. 大连交通大学, 2019.

[43]Wang C Y, Liao H Y M, Wu Y H, et al. CSPNet: A new backbone that can enhance learning capability of CNN[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. 2020: 390-391.

[44]Carion N, Massa F, Synnaeve G, et al. End-to-end object detection with transformers[C]//Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part I 16. Springer International Publishing, 2020: 213-229.

[45]Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017, 30.

[46]Yue X, Sun S, Kuang Z, et al. Vision transformer with progressive sampling[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021: 387-396.

[47]Liu Z, Lin Y, Cao Y, et al. Swin transformer: Hierarchical vision transformer using shifted windows[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021: 10012-10022.

[48]Cao Z, Simon T, Wei S E, et al. Realtime multi-person 2d pose estimation using part affinity fields[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017: 7291-7299.

[49]卫少洁,周永霞.基于轮廓关键点和LSTM的摔倒检测方法[J].计算机应用与软件,2022,39(04):213-217+241.

[50]Yu Y, Si X, Hu C, et al. A review of recurrent neural networks: LSTM cells and network architectures[J]. Neural Computation, 2019, 31(7): 1235-1270.

[51]张建军.基于手扶电梯监控视频的危险行为检测及研究[D].安徽大学,2021.

[52]邵延华,张铎,楚红雨,张晓强,饶云波.基于深度学习的YOLO目标检测综述[J].电子与信息学报, 2022, 44(10): 3697-3708.

[53]付苗苗,邓淼磊,张德贤.基于深度学习和Transformer的目标检测算法[J].计算机工程与应用,2023,59(01):37-48

[54]Qiu S, Xu X, Cai B. FReLU: flexible rectified linear units for improving convolutional neural networks[C]//2018 24th International Conference on Pattern Recognition (icpr). IEEE, 2018: 1223-1228.

[55]Woo S, Park J, Lee J Y, et al. CBAM: Convolutional block attention module[C]// European Conference on Computer Vision (ECCV). Germany: Springer, 2018: 3-19.

[56]Zhang Y, Xie F, Huang L, et al. A lightweight one-stage defect detection network for small object based on dual attention mechanism and PAFPN[J]. Frontiers in Physics, 2021, 9: 708097.

[57]Zheng Z, Wang P, Ren D, et al. Enhancing geometric factors in model learning and inference for object detection and instance segmentation[J]. IEEE Transactions on Cybernetics, 2021, 52(8): 8574-8586.