[1]方学宠,李 敏,白植志,等.基于改进RTMDet算法的自动扶梯行人异常行为检测[J].机械与电子,2026,44(07):31-38.
　FANG Xuechong,LI Min,BAI Zhizhi,et al.Abnormal Behavior Detection of Escalator Pedestrians Based on the Improved RTMDet Algorithm[J].Machinery & Electronics,2026,44(07):31-38.
点击复制

基于改进RTMDet算法的自动扶梯行人异常行为检测()
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
44
期数:
2026年07期
页码:
31-38
栏目:
智能检测
出版日期:
2026-07-25

文章信息/Info

Title:
Abnormal Behavior Detection of Escalator Pedestrians Based on the Improved RTMDet Algorithm
文章编号:
1001-2257(2026)07-0031-08
作者:
方学宠1; 李 敏1; 白植志1; 吴天琪1; 孙小清1; 陈向俊2; 易灿灿3
1.温州市特种设备检测科学研究院,浙江 温州 325000;
2.浙江省特种设备科学研究院,浙江 杭州 310000;
3.武汉科技大学机械传动与制造工程湖北省重点实验室,湖北 武汉 430081
Author(s):
FANG Xuechong1; LI Min1; BAI Zhizhi1; WU Tianqi1; SUN Xiaoqing1; CHEN Xiangjun2; YI Cancan3
(1.Wenzhou Special Equipment Inspection and Research Institute,Wenzhou 325000,China;
2.Zhejiang Special Equipment Research Institute,Hangzhou 310000,China;
3.Hubei Key Laboratory of MechanicalTransmission and Manufacturing Engineering,Wuhan University of Science and Technology,Wuhan 430081,China)
关键词:
自动扶梯; 异常行为检测; RTMDet; 注意力机制; 多尺度特征融合
Keywords:
escalator abnormal behavior detection; RTMDet; attention mechanism; multi-scale feature fusion
分类号:
TP391.4;TH236
文献标志码:
A
摘要:
针对自动扶梯场景中跌倒、蹲姿和携带行李等异常行为检测任务,现有方法存在背景干扰强、小目标检测困难、目标遮挡严重以及类别分布不均衡等问题,难以兼顾检测精度与实时性。为此,以RTMDet为基础框架,提出了一种改进的自动扶梯异常行为检测模型。在主干网络中引入CPCA 注意力机制,以增强模型对关键区域和有效特征的提取能力;在颈部网络中嵌入ASFM 自适应空间特征融合模块,以提升多尺度特征融合效果,增强对小目标和遮挡目标的表征能力;同时对损失函数进行改进,以缓解类别不平衡问题并提高边界框回归精度。实验结果表明,所提出的改进RTMDet算法在自建自动扶梯异常行为数据集上取得了较好的检测性能,Precision为91.8%,Recall为91.2%,mAP为94.9%,FPS为48.1 帧/s。与基线模型RTMDet相比,改进模型的mAP提高了4.8百分点,表明该方法能够在保证实时性的同时有效提升自动扶梯异常行为检测的准确性。
Abstract:
For abnormal behavior detection tasks in escalator scenes,such as falling,squatting,and carrying luggage,existing methods still face problems including severe background interference,difficulties in detecting small objects,heavy target occlusion,and imbalanced category distribution.Existing methods struggle to balance detection accuracy and real-time performance.To address these issues,an improved abnormal behavior detection model for escalators is proposed based on RTMDet framework.Specifically, the CPCA attention mechanism is incorporated into the backbone network to enhance the model’s ability to extract features from critical regions and informative cues.An Adaptive Spatial Feature Fusion Module(ASFM) is embedded into the neck network to improve multi-scale feature fusion and strengthen the representation of small and occluded targets.In addition,the loss function is refined to alleviate the problem of category imbalance and improve bounding box regression accuracy.Experimental results show that the proposed method achieves good detection performance on a self-built escalator abnormal behavior dataset,with a Precision of 91.8%,Recall of 91.2%,mAP of 94.9%,and FPS of 48.1.Compared with the baseline RTMDet,the improved model increases mAP by 4.8 percentage points,indicating that the proposed method can effectively improve the accuracy and robustness of escalator abnormal behavior detection while maintaining real time performance.

参考文献/References:

[1] 侯颖,胡鑫,赵瑞瑞,等.感兴趣区域YOLOBFROI的扶梯乘客安全检测算法[J].计算机工程与应用,2025,61(6):84-95.

[2] 侯颖,杨林,胡鑫,等.基于SwinT-YOLOX模型的自动扶梯行人安全检测算法[J].计算机工程,2024,50(3):277-289.
[3] 侯湘煜,胡华,方勇,等.基于贝叶斯网络的地铁车站楼扶梯运营系统风险评估模型[J].城市轨道交通研究,2025,28(6):238-244.
[4] 田伦,李国林,齐贺瑾妍.轨道交通自动扶梯智能化运维管理系统研究[J].机电信息,2026(5):38-41.
[5] 夏小松,唐川,赵再友,等.基于改进YOLOv5s的自动扶梯跌倒行为研究[J].中国特种设备安全,2025,41(增刊1):29-33.
[6] Wastupranata L M,Kong S G,Wang Lipo.Deep learning for abnormal human behavior detection in surveillance videos:a survey [J].Electronics,2024,13 (13):2579.
[7] Liu Shuoyan,Li Chao,Liu Yuxin,et al.SH-YOLO: small target high performance YOLO for abnormal behavior detection in escalator scene[J].IEICE Transactions on Information and Systems,2024,E107-D(11):1468-1471.
[8] Wang Qibing,Wang Dongyang,Lu Jiawei,et al.SAL-YOLO-DeepSeek:a lightweight real-time detection and LLM-driven decision framework for intelligent escalator safety monitoring[J].Scientific Reports, 2025,15:40600.
[9] Zhu Yongxuan,Nakamura H.The research on dangerous behavior recognition and warning system for escalators based on deep learning models[C]∥ ITE Technical Report,2024,48(29):1-5.
[10] Cao Zhe,Hidalgo G,Simon T,et al.OpenPose:realtime multi-person 2D pose estimation using part affinity fields[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2021,43(1):172-186.
[11] Sun Ke,Xiao Bin,Liu Dong,et al.Deep High-resolution representation learning for human pose estimation[ C]∥2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).New York: IEEE,2019:5693-5703.
[12] Fang Haoshu,Li Jiefeng,Tang Hongyang,et al.AlphaPose: whole-body regional multi-person pose estimation and tracking in real-time[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023,45(6):7157-7173.
[13] Maji D,Nagori S,Mathew M,et al.YOLO-Pose:enhancing YOLO for multi person pose estimation using object keypoint similarity loss[PP/OL].Version 1.arXiv(2022 04-14)[2026-05-22].https:∥arxiv.org/abs/2204.06806.
[14] Yan Sijie,Xiong Yuanjun,Lin Dahua.Spatial temporal graph convolutional networks for skeleton-based action recognition[C]∥ Thirty Second AAAI Conference on Artificial Intelligence,2018:7444-7452.
[15] Shi Lei,Zhang Yifan,Cheng Jian,et al.Two-stream adaptive graph convolutional networks for skeleton-based action recognition[C]∥2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).New York:IEEE,2019:12026-12035.
[16] Chen Yuxin,Zhang Ziqi,Yuan Chunfeng,et al.Channel-wise topology refinement graph convolution for skeleton-based action recognition[C]∥2021 IEEE/CVF International Conference on Computer Vision (ICCV).New York:IEEE,2021:13359-13368.
[17] Chen Weiming,Jiang Zijie,Guo Hailin,et al.Fall detection based on key points of human-skeleton using openpose[J].Symmetry,2020,12(5):744.
[18] Lyu Chengqi,Zhang Wenwei,Huang Haian,et al. RTMDet:an empirical study of designing real-time object detectors[PP/OL].Version 2.arXiv(2022-12-16 [2026-05-22].https:∥arxiv.org/abs/2212.07784.
[19] 张玉宁,贾渊,陈越.改进RTMDet的SAR舰船检测算法[J].计算机工程与应用,2024,60(22):314-322.
[20] 宋天平.基于改进RTMDet的水稻种子缺陷检测方法[J].电大理工,2025(4):1-9.
[21] 张杨,程智宇,陈允降,等.注意力机制增强的输煤传送带异物检测[J].智能科学与技术学报,2025,7(2):268-276.
[22] Wang Ao,Chen Hui,Liu Lihao,et al.YOLOv10: real-time end-to-end object detection[C]∥ Advances in Neural Information Processing Systems 37 (NeurIPS 2024),2024,37:107984-108011.
[23] Huang Shihua,Lu Zhichao,Cun Xiaodong,et al. Deim:detr with improved matching for fast convergence[ C]∥ 2025 IEEE/CVF Conference on ComputerVision and Pattern Recognition(CVPR).New York:IEEE,2025:15162-15171.
[24] Kong Yanning,Shang Xiangfeng,Jia Shijie.Drone DETR:efficient small object detection for remote sensing image using enhanced RT-DETR model[J].Sensors,2024,24(17):5496.

相似文献/References:

[1]孟庆宇.基于 ＥＥＭＤ- ＳＶＭ 的自动扶梯主驱动轴轴承故障诊断方法研究[J].机械与电子,2020,(05):51.
　Research on Fault Diagnosis Method for Main Drive Shaft Bearing of Escalator Based on EEMD-SVM [J].Machinery & Electronics,2020,(07):51.
[2]李自翔,张 强,范 豫.基于模态分量的自动扶梯轴承振动故障信号特征提取方法[J].机械与电子,2024,42(11):76.
　LI Zixiang,ZHANG Qiang,FAN Yu.Feature Extraction Method of Escalator Bearing Vibration Fault Signal Based on Modal Component[J].Machinery & Electronics,2024,42(07):76.

备注/Memo

备注/Memo:
收稿日期:2026-04-22
基金项目:国家自然科学基金资助项目(51805382);温州市科学技术局基础性公益科研项目(G20240066)
作者简介:方学宠 (1981-),男,浙江温州人,硕士,高级工程师,研究方向为特种设备检验;白植志 (1979-),男,浙江温州人,硕士,工程师,研究方向为特种设备检验,通信作者,E-mail:1600641596@qq.com。
更新日期/Last Update: 2026-08-27