自动扶梯在公共场合被广泛使用，传统扶梯巡检式日常管理消耗较多人力，庞大的 客流量时常会产生潜在的安全隐患，然而乘客摔倒事故却难以及时被发现，无法终止扶 梯运行，容易造成重大人身伤害,自动扶梯智能化监控成为预防事故发生的重要手段，为 市民打造更安全放心的乘梯环境。 （1）自动扶梯运行环境较复杂，行人较多，乘客尺度不断变化，远距离的小目标乘 客检测容易造成漏检和错检问题，本文提出一种基于感兴趣区域的轻量化改进YOLOv8 自动扶梯乘客摔倒检测算法。改进算法设计了感兴趣区域 BiFormer_ROI 注意力机制模 块，融合骨干网络可以有效屏蔽非扶梯背景区域的复杂环境干扰，有效提高小目标的检 测率。考虑实际工业应用需要，Neck网络采用GhostSlimPAFPN轻量化模型，在保持检 测精度的同时有效减少模型参数量。此外，采用具有目标尺寸自适应惩罚因子的PIoU v2 损失函数改进 Head 网络，从而实现更快的收敛和更高的检测精度。针对自动扶梯乘客 摔倒无公开数据集问题，分别在商场、地铁、机场、火车站和医院等场所通过模拟摔倒 动作采集了425段视频序列，从中提取11850张关键帧图像，同时通过网络收集了3500 幅扶梯乘客真实摔倒图像，最终建成包含15350张图像的图像样本库。在自建扶梯乘客 摔倒数据集上，实验结果显示本文改进算法检测性能明显提高，并有效减少误检和漏检 问题，乘客摔倒平均检测精度可以达到92.9%，检测帧率为87.7fps，具有良好的实时性， 可以更好地保障乘客安全乘梯。 （2）开发实现自动扶梯智能监控平台，采用TensorRT对本文改进算法进行模型优 化，并在NVIDIA Jetson Nano 嵌入式平台进行推理部署，同时使用PyQt5软件设计开发 前端平台界面。自动扶梯智能监控平台可以 24 小时全天候高效视频监控，系统采用本 文改进算法能够实时、精准地检测到扶梯乘客摔倒行为，同时发布语音播报，给管理人 员发送预警信息，并向扶梯控制系统发送紧急缓停的应急措施信号，从而保证乘客安全， 降低事故危害等级，显著减轻扶梯安全管理人员工作量。

Escalators are widely used in public places. The traditional daily management of escalator inspection consumes more manpower, and the huge passenger flow often produces potential safety hazards. However, passenger fall accidents are difficult to be discovered in time, and the operation of escalators cannot be stopped, which is easy to cause serious personal injuries. To create a safer and more secure riding environment for the public. (1) The escalator operating environment is complex, there are more pedestrians, and the passenger size is constantly changing, and the long-distance small-target passenger detection is easy to cause missing detection and wrong detection problems. A lightweight improved YOLOv8 escalator passenger fall detection algorithm based on the area of interest is proposed. The improved algorithm designed the BiFormer_ROI attention mechanism module of the region of interest. The fusion backbone network can effectively shield the complex environmental interference of the non-escalator background region, and effectively improve the detection rate of small targets.Considering the needs of actual industrial applications, Neck network adopts GhostSlimPAFPN lightweight model, which can effectively reduce the number of model parameters while maintaining the detection accuracy.In addition, PIoU v2 loss function with target size adaptive penalty factor is used to improve the Head network, thus achieving faster convergence and higher detection accuracy. In order to solve the problem of escalator passengers falling without public data set, 425 video sequences were collected in shopping malls, subways, airports, railway stations, hospitals and other places by simulated falling action, from which 11850 key frame images were extracted, and 3,500 real escalator passengers falling images were collected through the network. Finally, the image sample library containing 15,350 images was built. On the self-built escalator passenger fall data set, the experimental results show that the detection performance of the improved algorithm in this thesis is significantly improved, and the problem of false detection and missing detection is effectively reduced. The average detection accuracy of passenger fall can reach 92.9%, and the detection frame rate is 87.7fps, which has good real-time performance and can better guarantee the safety of passengers taking the escalator. (2) Developed and realized the escalator intelligent monitoring platform, adopted TensorRT to optimize the model of the improved algorithm in this thesis, and carried out inference deployment on the NVIDIA Jetson Nano embedded platform, and used PyQt5 software to design and develop the front-end platform interface. The intelligent escalator monitoring platform can provide 24-hour and all-weather video surveillance. The improved algorithm adopted in this thesis can detect the falling behavior of escalator passengers in real time and accurately. At the same time, the system can release voice broadcast, send early warning information to management personnel, and send emergency measures signals of emergency suspension to the escalator control system, so as to ensure the safety of passengers and reduce the accident hazard level. Significantly reduce the workload of escalator safety management personnel.

[1] Xing Y, Chen S, Zhu S, et al. Analysis factors that influence escalator-related injuries in metro stations based on bayesian networks: a case study in China[J]. International journal of environmental research and public health, 2020, 17(2): 481 1-21.

[2] Osipov, V., Zhukova, N., Subbotin, A. et al. Intelligent escalator passenger safety management[J]. Scientific Reports, 2022, 12: 5506 1-16.

[3] 陈旻.浅析自动扶梯及自动人行道中的“剪切”危险[J].机电技术,2009,32(04):104-107.

[4] Zou Z, Chen K, Shi Z, et al. Object detection in 20 years: A survey[J]. Proceedings of the IEEE, 2023.

[5] P . Viola and M. Jones, “Rapid object detection using a boosted cascade of simple features,” in Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit. (CVPR), Dec. 2001, pp.1–9.

[6] N. Dalal and B. Triggs, “Histograms of oriented gradients for human detection,” in Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., vol. 1, no. 1, Jun. 2005, pp.886–893.

[7] P . Felzenszwalb, D. McAllester , and D. Ramanan,“ A discriminatively trained, multiscale,deformable part model,” in Proc. IEEE Conf. Comput. Vis.Pattern Recognit., Jun. 2008,pp. 1–8.

[8] Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks[J]. Advances in neural information processing systems, 2012, 25.

[9] R. Girshick, J. Donahue, T . Darrell, and J. Malik,“Rich feature hierarchies for accurate object detection and semantic segmentation,” in Proc.IEEE Conf. Comput. Vis. Pattern Recognit.,Jun. 2014, pp. 580–587.

[10] K. He, X. Zhang, S. Ren, and J. Sun, “Spatial pyramid pooling in deep convolutional networks for visual recognition,” in Proc. ECCV. Cham,Switzerland: Springer , 2014, pp.346–361.

[11] R. Girshick, “Fast R-CNN,” in Proc. IEEE Int. Conf.Comput. Vis. (ICCV), Dec. 2015, pp. 1440–1448.

[12] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards real-time object detection with region proposal networks,” in Proc. Adv. Neural Inf. Process. Syst., 2015, pp. 91–99.

[13] T .-Y . Lin, P . Goyal, R. Girshick, K. He, and P . Dollar ,“Focal loss for dense object detection,” IEEE Trans.Pattern Anal. Mach. Intell., vol. 42, no. 2,pp. 318–327, Feb. 2020.

[14] He K, Gkioxari G, Dollár P, et al. Mask r-cnn[C]//Proceedings of the IEEE international conference on computer vision. 2017: 2961-2969.

[15] 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.

[16] Fu C Y, Liu W, Ranga A, et al. Dssd: Deconvolutional single shot detector[J]. arXiv preprint arXiv:1701.06659, 2017.

[17] Li Z, Zhou F. FSSD: feature fusion single shot multibox detector[J]. arXiv preprint arXiv:1712.00960, 2017.

[18] 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. 2017: 2980-2988.

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

[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] Redmon J, Farhadi A. YOLO9000: better, faster, stronger[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 7263-7271.

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

[23] Jocher G, Chaurasia A, Stoken A, et al. ultralytics/yolov5: v7. 0-yolov5 sota realtime instance segmentation[J]. Zenodo, 2022.

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

[25] 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.

[26] 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.

[27] Sun G, Wang Z. Fall detection algorithm for the elderly based on human posture estimation[C]//2020 Asia-Pacific Conference on Image Processing, Electronics and Computers (IPEC). IEEE, 2020: 172-176.

[28] M. J. Mathie, A. C. F. Coster, N. H. Lovell, et al., “Accelerometry: providing an integrated, practical method for long-term, ambulatory monitoring of human movement,” Physiological Measurement, vol. 25,no. 2, pp. R1-R20, 2004.

[29] M. Kangas, I. Vikman, J. Wiklander, et al., “Sensitivity and specificity of fall detection in people aged 40 years and over,” Gait & Posture, vol. 29, no. 4, pp. 0-574, 2009.

[30] Tamura T, Yoshimura T, Sekine M, et al. A wearable airbag to prevent fall injuries[J]. IEEE Transactions on Information Technology in Biomedicine, 2009, 13(6): 910-914.

[31] 薛源.基于多传感器的老人跌倒检测系统的研究与应用[D].武汉：武汉理工大学,2011.

[32] Litvak D, Zigel Y, Gannot I. Fall Detection of Elderly Through Floor Vibrations and Sound[J]. Annual international conference of the IEEE engineering in medicine and biology Society, 2008:4632-4635

[33] Rimminen H, J. Lindström, Linnavuo M, et al. Detection of Falls Among the Elderly by a Floor Sensor Using the Electric Near Field[J]. IEEE transactions on information technology in biomedicine: a publication of the IEEE Engineering in Medicine and Biology Society, 2010, 14(6):1475-1476.

[34] Zheng C, Wu W, Chen C, et al. Deep learning-based human pose estimation: A survey[J]. ACM Computing Surveys, 2023, 56(1): 1-37.

[35] Dubey S, Dixit M. A comprehensive survey on human pose estimation approaches[J]. Multimedia Systems, 2023, 29(1): 167-195.

[36] Sun G, Wang Z. Fall detection algorithm for the elderly based on human posture estimation[C]//2020 Asia-Pacific Conference on Image Processing, Electronics and Computers (IPEC). IEEE, 2020: 172-176.

[37] Pires S, Rodrigues S, Arokiadass L B, et al. A Real-Time Position Monitoring SystemfFor Fall Detection and Analysis Using Human Pose Estimation[C]//2021 4th Biennial International Conference on Nascent Technologies in Engineering (ICNTE). IEEE, 2021: 1-7.

[38] Ali R, Hutomo I S, Van L D, et al. A Skeleton-based View-Invariant Framework for Human Fall Detection in an Elevator[C]//2022 IEEE International Conference on Industrial Technology (ICIT). IEEE, 2022: 1-6.

[39] Guan Z, Li S, Cheng Y, et al. A video-based fall detection network by spatio-temporal joint point model on edge devices[C]//2021 Design, Automation & Test in Europe Conference & Exhibition (DATE). IEEE, 2021: 422-427.

[40] Liu C M, Huang Z S, Chen Y L. Fall Detection System for Depression Angle RGB Image based on Human Skeletonch[C]//2022 IET International Conference on Engineering Technologies and Applications (IET-ICETA). IEEE, 2022: 1-2.

[41] Gao M, Li J, Zhou D, et al. Fall detection based on OpenPose and MobileNetV2 network[J].IET Image Processing, 2023, 17(3): 722-732.

[42] Liu S, Tan R, Yan Z. Fall detection based on lightweight OpenPose algorithm[C]//Third International Conference on Intelligent Computing and Human-Computer Interaction (ICHCI 2022). SPIE, 2023, 12509: 663-670.

[43] Fei K, Wang C, Zhang J, et al. Flow-pose Net: An effective two-stream network for fall detection[J]. The Visual Computer, 2023, 39(6): 2305-2320.

[44] LUO B. Human Fall Detection for Smart Home Caring using Yolo Networks[J]. International Journal of Advanced Computer Science and Applications(IJACSA),2023,14(4).

[45] 杨雪旗,唐旭,章国宝等.基于 YOLO 网络的人体跌倒检测方法[J].扬州大学学报(自然科学版),2019,22(02):61-64+78.

[46] Goodfellow I, Bengio Y, Courville A. Deep learning[M]. MIT press, 2016.

[47] Hu J, Shen L, Sun G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 7132-7141.

[48] Hou Q, Zhou D, Feng J. Coordinate attention for efficient mobile network design[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021: 13713-13722.

[49] Zhu L, Wang X, Ke Z, et al. BiFormer: Vision Transformer with Bi-Level Routing Attention[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2023: 10323-10333.

[50] Li H, Li J, Wei H, et al. Slim-neck by GSConv: A better design paradigm of detector architectures for autonomous vehicles[J]. arXiv preprint arXiv:2206.02424, 2022.

[51] 高晗,田育龙,许封元,仲盛.深度学习模型压缩与加速综述[J].软件学报,2021(01):68-92.

[52] HAN K, WANG Y, TIAN Q, et al. Ghostnet: More features from cheap operations[C].Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020:1580-1589.

[53] Liu C, Wang K, Li Q, et al. Powerful-IoU: More straightforward and faster bounding box regression loss with a nonmonotonic focusing mechanism[J]. Neural Networks, 2024, 170: 276-284.

[54] 黄率.基于 PYQT5 的 AI 图像识别工具[J].现代工业经济和信息化,2023,13(01):90-91+94.