传感器世界2024，Vol.30Issue(5) ：32-37.DOI:10.16204/j.sw.issn.1006-883X.2024.05.007

基于多传感器融合的行人异常行为识别方法

Identification Method of Pedestrian Anomaly Behavior Based on Multi Sensor Fusion

赵莉苹 李辉
传感器世界2024，Vol.30Issue(5) ：32-37.DOI:10.16204/j.sw.issn.1006-883X.2024.05.007
✕
  • NETL
  • NSTL
  • 万方数据

基于多传感器融合的行人异常行为识别方法

Identification Method of Pedestrian Anomaly Behavior Based on Multi Sensor Fusion

赵莉苹 1李辉2
扫码查看

作者信息

  • 1. 郑州科技学院信息工程学院,河南郑州 450064
  • 2. 河南思维信息技术有限公司,河南郑州 450000
  • 折叠

摘要

常规的行人异常行为识别方法无法针对行人行为特征进行识别,导致识别准确性较低,因此,设计了一种基于多传感器融合的行人异常行为识别方法.在多传感器融合中,通过摄像头得到人体不同行为的最小外接矩形框,并计算出人的宽高比,以反映行人的运行特征;然后根据平行式双目视觉测量原理,获取行人位置信息,并利用空间对齐的方式,获得行人位置信息,实现行人异常行为点云信息与图像信息一致的目标,从而提高信息融合的精准度;最后结合行人的步长、速度、方向等轨迹数据对行人异常行为进行分类,并通过奖励或惩罚机制优化分类识别结果,完成行人异常行为识别.实验结果表明,使用该方法进行行人异常行为识别时,行人异常行为识别的平均精确度在0.98～1.00的范围内变化,行人异常行为识别的均值平均精度在0.97～1.00的范围内变化.由此表明,该方法的识别准确性更高,能够应用于实际生活中.

Abstract

Conventional methods for recognizing pedestrian anomalous behavior fail to capture running characteristics,resulting in low recognition accuracy.As a result,a Multi-sensor Fusion based approach has been developed for recognizing pedestrian anomalous behavior.Within the context of multi sensor fusion,a minimal outer rectangular frame capturing various human behaviors is obtained from the camera,and the aspect ratio of the individuals is calculated to reflect pedestrian running characteristics.Subsequently,leveraging the principles of parallel binocular vision measurement,the system obtains location information of pedestrians.The method of spatial alignment is employed to ensure consistency between the point cloud information of pedestrian abnormal behavior and the image information,thereby enhancing the precision of information fusion.Finally,abnormal pedestrian behavior is classified based on track data such as step size,speed,and direction.Classification and recognition results are then optimized using a reward or penalty mechanism,completing the identification of pedestrian abnormal behavior.Experimental results demonstrate that the average accuracy of recognizing pedestrian abnormal behavior ranges from 0.98 to 1.00,and the average accuracy of classifying pedestrian abnormal behavior varies from 0.97 to 1.00.These results indicate that the proposed method offers heightened recognition accuracy and practical application potential.

关键词

多传感器融合/行人/空间对齐/异常行为/分类/识别方法

Key words

multi sensor fusion/pedestrians/spatial alignment/abnormal behavior/classification/identification method

引用本文复制引用

基金项目

郑州科技学院校级教改项目(2022JGYB16)

出版年

2024
传感器世界
北京信息科技大学

传感器世界

影响因子：0.196
ISSN：1006-883X
段落导航相关论文