航站楼集群安全识别与预警关键技术研究

发布时间：2018-06-20 08:33

本文选题：航站楼集群安全 + 人数统计　；参考：《南京航空航天大学》2014年博士论文

【摘要】：随着我国民航运输业的快速发展,机场旅客数量也随之剧增,由此导致的大量旅客群体性事件给机场的安全生产带来了很大压力。尽管现代机场都已装备了视频监控系统以辅助管理者及时发现航站楼中的异常情况,但传统的视频监控系统还停留在人工观测的预警方式,极为耗费人力。随着计算机数字图像技术的不断发展,新一代智能化视频监控系统正成为新的研究热点。它能够利用图像处理和模式识别算法自主性地分析视频数据,并对其中发生的异常情况给出预警,从而减轻安全监管过程中的人力消耗。为此,本文在深入分析国内外智能视频监控技术研究现状以及我国机场航站楼安全管理需求的基础上,重点研究了航站楼集群安全识别与预警问题中所涉及的若干关键技术。首先,本文对智能视频处理的基础技术前景提取进行了研究,设计了一种改进的高斯混合背景模型用于前景提取。改进后的模型将传统模型中的标准差用平均偏差代替,节约了计算量。同时,还通过引入HSV阴影滤除模型减少了阴影对前景提取的影响。此外,设计了一种归一化前景面积计算方法,以解决前景面积这一传统图像特征在用于人数统计时由于透视效应的影响而无法准确估计人数的问题。其次,本文从人群密度估计的角度对旅客人数统计问题展开了研究,分别提出了三种不同类型的人数统计方法。第一种算法通过一个具有尺寸自适应特性的滑动窗和一组判别条件在对前景二值图进行行人检测扫描的基础上实现了中低密度场景下的人数统计。第二种算法基于归一化前景和角点信息特征设计一种遮挡因子,以克服遮挡对人数统计效果的影响,实现了中高人群密度场景下的人数统计。第三种算法基于单应性原理设计一种考虑了遮挡因素的目标匹配策略,以提升单视点场景中人数统计效果,实现了多视点条件下的人数统计。另外,针对航站楼集群安全管理中所关注的旅客异常行为问题,本文从行人异常行为识别角度,设计了两种算法,分别用于识别旅客聚集行为和旅客肢体冲突行为。第一种算法在计算出归一化前景面积和二维联合熵的基础上,设计了一种人群聚集检测参数,以实现对人群聚集行为的有效识别。第二种算法在计算出的前景区域光流信息基础上,利用提取的光流方向熵特征,实现了对行人肢体冲突行为的有效识别。最后,本文设计了一种航站楼集群安全识别与预警原型系统。它能在获取传统视频监控系统中视频数据的基础上,利用相关图像处理算法和并行处理技术,实现多路视频条件下的人数统计和行人异常行为识别等功能,并进一步验证了本文所设计的旅客人数统计和旅客异常行为识别算法的有效性。
[Abstract]:With the rapid development of China's civil aviation transportation industry, the number of airport passengers has also increased dramatically, resulting in a large number of passenger group incidents to bring great pressure to airport safety production. Although modern airports have been equipped with video surveillance systems to help managers discover the abnormal situation in the terminal building in time, the traditional video surveillance system still stays in the early warning mode of manual observation, which is extremely labor-intensive. With the development of computer digital image technology, a new generation of intelligent video surveillance system is becoming a new research hotspot. It can make use of image processing and pattern recognition algorithms to analyze video data autonomously, and give early warning of abnormal situation in it, thus reducing the manpower consumption in the process of security supervision. Therefore, on the basis of deeply analyzing the research status of intelligent video surveillance technology at home and abroad and the security management requirements of airport terminal building in China, this paper focuses on some key technologies involved in the identification and early warning of terminal cluster security. Firstly, this paper studies the basic technology of intelligent video processing, and designs an improved Gao Si hybrid background model for foreground extraction. The improved model replaces the standard deviation of the traditional model with the average deviation, which saves the calculation cost. At the same time, the influence of shadow on foreground extraction is reduced by introducing HSV shadow filtering model. In addition, a method of calculating normalized foreground area is designed to solve the problem that foreground area, a traditional image feature, can not be estimated accurately because of the influence of perspective effect. Secondly, from the point of view of crowd density estimation, this paper studies the problem of passenger statistics, and puts forward three different types of statistics methods. In the first algorithm, a sliding window with adaptive size and a set of discriminant conditions are used to realize the population statistics of low and low density scenes on the basis of the pedestrian detection scan of the foreground binary map. The second algorithm designs an occlusion factor based on normalized foreground and corner information features to overcome the effect of occlusion on the population statistics and realize the population statistics under the scenario of high population density. The third algorithm is based on the principle of homography to design a target matching strategy which takes into account occlusion factors to improve the effect of population statistics in single view scene and realize the population statistics under the condition of multiple viewpoints. In addition, aiming at the problem of abnormal behavior of passengers concerned in the cluster security management of terminal building, this paper designs two algorithms from the perspective of pedestrian abnormal behavior recognition, which are used to identify passenger aggregation behavior and passenger limb conflict behavior respectively. On the basis of computing normalized foreground area and two-dimensional joint entropy, the first algorithm designs a crowd aggregation detection parameter to realize the effective identification of crowd aggregation behavior. On the basis of the calculated optical flow information in foreground region, the second algorithm uses the extracted entropy feature of optical flow direction to realize the effective identification of pedestrian limb conflict behavior. Finally, a prototype system of terminal cluster security identification and early warning is designed. On the basis of obtaining video data in traditional video surveillance system, it can use correlation image processing algorithm and parallel processing technology to realize the functions of number statistics and pedestrian abnormal behavior identification under multi-channel video condition. Furthermore, the effectiveness of the proposed algorithm is verified.
【学位授予单位】：南京航空航天大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：V351.3 ;TN948.6

【参考文献】