基于智能视频分析的小区异常事件检测

发布时间：2018-05-13 20:46

本文选题：行人徘徊检测 + 围栏攀爬检测　；参考：《海南大学》2017年硕士论文

【摘要】：小区的安防保障着我们的生活质量,随着计算机技术的发展,将智能视频分析技术运用在小区监控系统中,具有省时省力的作用。自动对视频监控画面进行运动目标检测、跟踪和行为判断是当前智能视频分析技术非常热门的研究内容。本文总结和比较了几种经典的己有算法,并重点研究了三种小区内异常事件的检测方法,如行人徘徊检测、围栏攀爬检测和物品遗留与搬移检测。本文的主要工作可归纳以下五个方面:(1)研究了帧差法、背景差分法和光流法三种运动目标检测算法,并在公共数据集和自有数据集上进行实验对比。由于行人在小区监控视频中的重要性,又重点研究了三种行人检测的方法,并进行比较。(2)研究了均值漂移跟踪算法、压缩跟踪算法和基于时空上下文的跟踪算法三种经典的运动目标跟踪算法,并对这三种算法在公共数据集和自有数据集上进行实验对比。(3)基于上述两部分的先验工作,提出了在感兴趣区域内进行行人检测再跟踪,最后基于运动轨迹的位移与路程差来判断是否有行人徘徊事件发生的方法,并验证了方法的有效性。(4)基于攀爬姿势的独特性,采用HOG-LBP的联合特征,提出了对围栏中上方的监控画面进行二分类检测来判断是否有围栏攀爬事件发生的方法,并验证了方法的有效性。(5)基于第一部分的先验工作,提出了使用不同速率建立两个混合高斯模型从而来提取在一定时间内运动状态发生改变的非背景物体,对该物体进行行人检测,排除为停留的行人后再根据该物体与背景模型的关系,从而判断物品被遗留还是被搬移的方法,并验证了方法的有效性。
[Abstract]:With the development of computer technology, the application of intelligent video analysis technology in community monitoring system has a time-saving and labor-saving role. Automatic moving target detection, tracking and behavior judgment for video surveillance images are very popular research contents of intelligent video analysis technology. This paper summarizes and compares several classical self-existing algorithms, and focuses on three detection methods of abnormal events in residential areas, such as pedestrian hovering detection, fence climbing detection, and the detection of items left behind and moved. The main work of this paper can be summarized as follows: 1) three kinds of moving target detection algorithms, frame difference method, background difference method and optical flow method, are studied, and the experimental results are compared between the common data set and the own data set. Because of the importance of pedestrian in the video surveillance of residential area, three pedestrian detection methods are studied, and the mean drift tracking algorithm is studied. Compression tracking algorithm and spatio-temporal context based tracking algorithm are three classical moving target tracking algorithms, and the three algorithms are compared with each other on the common data set and the own data set. This paper proposes a method to detect and track pedestrians in the region of interest. Finally, based on the displacement and distance difference of motion trajectory, the method is proposed to judge whether there is a pedestrian hovering event or not, and the validity of the method is verified. 4) based on the uniqueness of climbing posture, a new method is proposed to determine whether there is a pedestrian hovering event or not. Based on the joint features of HOG-LBP, this paper presents a two-class method to detect the monitoring images above the fence to determine whether there is a climbing event in the fence, and verifies the validity of the method based on the priori work of the first part. In this paper, two mixed Gao Si models with different rates are proposed to extract the non-background objects whose motion states change within a certain time, and to detect the objects by pedestrian detection. According to the relationship between the object and the background model, the method of whether the object is left behind or moved is judged, and the validity of the method is verified.
【学位授予单位】：海南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】