基于骨架识别的城轨车站监控视频乘客行为特征辨识研究  

Passenger Behavior Feature Identification in Urban Rail Station Surveillance Videos Using Skeleton Recognition Techniques

作　　者：管洋 贾利民[1] 陶思涵 豆飞 GUAN Yang;JIA Limin;TAO Sihan;DOU Fei(School of Traffic and Transportation,Beijing Jiaotong University,Beijing 100044;CRSC Research&Design Institute Group Co.,Ltd.,Beijing 100070;Beijing Union University,Beijing 100101)

机构地区：[1]北京交通大学交通运输学院,北京100044 [2]北京全路通信信号研究设计院集团有限公司,北京100070 [3]北京联合大学,北京100101

出　　处：《都市快轨交通》2025年第1期106-111,共6页Urban Rapid Rail Transit

基　　金：北京市自然科学基金-丰台轨道交通前沿研究联合基金资助(L221006)。

摘　　要：城市轨道交通领域传统监控分析方法对视频监控图像(如摔倒、晕倒和打斗等异常行为识别)漏识率高、参数调整复杂,且难以高效地应用于现实城轨车站监控场景,针对此问题,采用基于骨架模式识别的人体姿态特征辨识框架,引入基于人体骨架的姿态估计技术,采用Alpha Pose模型对乘客姿态进行精确估计,并结合时空图卷积网络(spatial temporal graph convolutional networks,ST-GCN)模型的方法,实现对城轨车站监控场景中异常行为的辨识。在COCO数据集和MPII数据集上分别达到了72.3 mAP和82.1 mAP的效果,相比较于Open Pose模型提升高达17%,验证了模型的有效性和实用性。结果表明,本文所提出的方法不仅提高了乘客行为的识别速度,同时具备对复杂场景的适应能力,为城轨安全监控提供一种新的技术方案。In order to solve the problem that the traditional monitoring analysts in the field of urban rail transit have a high false-negative rates and complex parameter adjustment of abnormal behaviors such as falling,fainting and fighting,making them difficult to apply efficiently to actual urban rail station monitoring scenarios,this paper proposes a human posture feature recognition framework based on skeleton pattern recognition,introducing the attitude estimation technology based on human skeleton.The Alpha Pose model is used to accurately estimate the posture of passengers,and combined with the Spatial Temporal Graph Convolutional Networks model,it achieves abnormal behavior recognition in the monitoring scenario of urban rail stations.By achieving 72.3 mAP on the COCO dataset and 82.1 mAP on the MPII dataset,the performance is improved by up to 17%compared to the OpenPose model,verifying the effectiveness and practicality of the model.The results show that the method proposed in this paper not only improves the recognition speed of passenger behavior but also has the ability to adapt to complex scenarios,providing a new technical solution for urban rail safety monitoring.

关 键 词：轨道交通 骨架识别 模式识别 城轨车站安全 乘客行为特征辨识 ST-GCN 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]