视频监控系统下航站楼旅客异常行为检测方法研究
发布时间：2018-11-18 21:11
【摘要】：机场航站楼作为民航交通和运输的重要枢纽之一,旅客吞吐量较高,属于异常行为多发区域。而传统的航站楼监控系统因监控点多且需人工监测等特点显得非常耗时费力,已很难满足机场安全管理的需要。因此,采用智能视频分析技术,主动对航站楼内旅客异常行为进行实时检测并报警,能有效协助机场安保人员处理异常事件,完善机场对突发事件的快速反应能力。本文主要对智能监控下旅客群聚、奔跑以及遗留物体三种异常行为的检测方法所涉及的关键技术进行深入研究,主要内容有:首先,视频监控下运动目标的提取。在传统高斯混合模型的基础上,优化高斯模型的均值及方差的更新机制,引入HSV空间阴影去除方法,从而克服初始建模速度慢且存在大量阴影的缺点,较好地重建视频图像的背景模型,实现运动目标的提取。其次,旅客群聚和奔跑异常行为检测,根据这两种异常行为在人群密度和运动特征上的表现形式的差异,提出各自的判断指标对异常行为进行检测。采用摄像机透视效应的加权前景面积以及前景的二维联合信息熵设计人群密度指标;利用金字塔LK光流特征计算能量和加权方向直方图熵对运动特征进行定量描述。通过不同模拟机场航站楼场景下的视频序列进行测试,对本文提出算法的可靠性进行验证。最后,基于遗留物检测的旅客异常行为检测。采用不同更新速率的改进GMM模型对场景的双重背景进行建模,去除行人等运动目标的干扰,实现对场景中短期静止目标的提取;根据目标的多个特征对其进行跟踪分析,当其在场景中停留时间超过设定的阈值,则将其判断为遗留物;结合遗留物的状态变化和历史图像信息,判断旅客是否为滞留物体或丢失物体的异常行为。通过不同模拟机场航站楼场景下的视频序列进行测试,验证本文算法的准确性。
[Abstract]:As one of the important hubs of civil aviation traffic and transportation, airport terminal has a high passenger throughput and belongs to a region with frequent abnormal behavior. The traditional terminal monitoring system is very time-consuming and laborious because of the many monitoring points and the need of manual monitoring. It is difficult to meet the needs of airport security management. Therefore, using intelligent video analysis technology to detect and alarm the abnormal behavior of passengers in terminal building in real time can effectively assist airport security personnel to deal with abnormal events and improve the ability of airport to respond quickly to emergencies. In this paper, the key technologies involved in the detection of passenger clustering, running and residual objects under intelligent surveillance are studied. The main contents are as follows: first, the extraction of moving targets under video surveillance. On the basis of the traditional Gao Si mixed model, the updating mechanism of the mean and variance of Gao Si model is optimized, and the HSV space shadow removal method is introduced to overcome the shortcomings of slow initial modeling speed and a large number of shadows. The background model of video image is reconstructed well and the moving object is extracted. Secondly, the abnormal behavior of passenger clustering and running is detected. According to the difference of the two abnormal behaviors in the density and movement characteristics of the crowd, the paper puts forward their own judgment indexes to detect the abnormal behavior. The weighted foreground area of the camera perspective effect and the two-dimensional joint information entropy of the foreground are used to design the population density index, and the calculated energy and weighted direction histogram entropy of the pyramid LK optical flow feature are used to quantitatively describe the motion feature. The reliability of the proposed algorithm is verified by testing the video sequences in different simulated airport terminal scenarios. Finally, the passenger abnormal behavior detection based on the residue detection. An improved GMM model with different updating rates is used to model the dual background of the scene, to remove the interference of moving objects such as pedestrians, and to achieve the extraction of the short-term still targets in the scene. According to the multiple features of the target, the target is tracked and analyzed, and when the residence time in the scene exceeds the set threshold, the target is judged as a remnant. Combined with the state change of the remnant and the historical image information, the abnormal behavior of the passenger is judged whether the passenger is a stranded object or a lost object. The accuracy of this algorithm is verified by testing the video sequences of different simulated airport terminal scenarios.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;TN948.6
，
本文编号：2341251
[Abstract]:As one of the important hubs of civil aviation traffic and transportation, airport terminal has a high passenger throughput and belongs to a region with frequent abnormal behavior. The traditional terminal monitoring system is very time-consuming and laborious because of the many monitoring points and the need of manual monitoring. It is difficult to meet the needs of airport security management. Therefore, using intelligent video analysis technology to detect and alarm the abnormal behavior of passengers in terminal building in real time can effectively assist airport security personnel to deal with abnormal events and improve the ability of airport to respond quickly to emergencies. In this paper, the key technologies involved in the detection of passenger clustering, running and residual objects under intelligent surveillance are studied. The main contents are as follows: first, the extraction of moving targets under video surveillance. On the basis of the traditional Gao Si mixed model, the updating mechanism of the mean and variance of Gao Si model is optimized, and the HSV space shadow removal method is introduced to overcome the shortcomings of slow initial modeling speed and a large number of shadows. The background model of video image is reconstructed well and the moving object is extracted. Secondly, the abnormal behavior of passenger clustering and running is detected. According to the difference of the two abnormal behaviors in the density and movement characteristics of the crowd, the paper puts forward their own judgment indexes to detect the abnormal behavior. The weighted foreground area of the camera perspective effect and the two-dimensional joint information entropy of the foreground are used to design the population density index, and the calculated energy and weighted direction histogram entropy of the pyramid LK optical flow feature are used to quantitatively describe the motion feature. The reliability of the proposed algorithm is verified by testing the video sequences in different simulated airport terminal scenarios. Finally, the passenger abnormal behavior detection based on the residue detection. An improved GMM model with different updating rates is used to model the dual background of the scene, to remove the interference of moving objects such as pedestrians, and to achieve the extraction of the short-term still targets in the scene. According to the multiple features of the target, the target is tracked and analyzed, and when the residence time in the scene exceeds the set threshold, the target is judged as a remnant. Combined with the state change of the remnant and the historical image information, the abnormal behavior of the passenger is judged whether the passenger is a stranded object or a lost object. The accuracy of this algorithm is verified by testing the video sequences of different simulated airport terminal scenarios.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;TN948.6
，
本文编号：2341251
本文链接：https://www.wllwen.com/kejilunwen/wltx/2341251.html
