高速公路交通异常事件检测算法研究

发布时间：2018-06-08 19:30

本文选题：阴影去除 + Kalman滤波算法　；参考：《兰州交通大学》2017年硕士论文

【摘要】：近年来,随着计算机存储和运算能力的不断提高,人工智能、模式识别技术的迅猛发展,基于视频的交通事件检测技术成为智能交通领域研究的热点问题。交通事件自动检测系统是交通视频监控系统智能化和自动化的关键,为快速处理交通事件、减少交通延误、避免二次交通事故的发生提供条件,为高速公路运营管理提供了新的突破口。但如何高效、准确、快速地实现交通事件自动检测,仍是当前智能交通领域面对的一大难题。本文从实际应用出发,以高速公路视频序列为研究对象,从运动目标检测、跟踪和异常行为描述等几个关键技术着手进行研究,设计了高速公路逆行、停车、变道异常事件自动检测算法。本文对上述三种异常事件的研究主要包括以下几个方面的内容:在运动目标检测方面,采用均值法建立背景模型,以背景差法提取运动目标前景;针对存在阴影的运动目标前景,提出了一种基于边缘和HSV颜色空间相结合的方法去除阴影,并结合形态学处理方法提取出完整的运动目标前景,为有效的运动目标跟踪提供了基础。在运动目标跟踪方面,以车辆的质心和面积为基本特征对车辆进行跟踪,结合Kalman滤波算法寻求运动目标特征的最优估计,利用欧式距离计算运动目标的位置距离和面积大小差异寻找最佳匹配完成运动目标的跟踪;针对车辆间遮挡会使跟踪目标丢失的现象,本文提出了面积筛选的方法用不同的方式对车辆进行跟踪,最终获得车辆的运动轨迹,为异常事件的判断提供了依据。在异常事件检测方面,通过分析车辆的运动轨迹可以直观的了解车辆的运动方向,将车辆的运动方向与道路规定的正方向进行比较判断车辆逆行事件;通过分析车辆的运动轨迹可以间接获得车辆的瞬时速度、加速度、质心位置变化等交通参数,分析这些交通参数的变化判断车辆是否发生违章停车事件;通过分析车辆运动轨迹与基准车道线间距离的离散程度判断车辆是否发生变道事件。本文对不同路段高速公路实际交通视频序列进行测试,实验结果验证了本文异常事件自动检测算法行之有效,能够准确的检测出逆行、停车、变道异常事件,具有很好的实用性。
[Abstract]:In recent years, with the continuous improvement of computer storage and computing ability, artificial intelligence, pattern recognition technology, the rapid development of video-based traffic incident detection technology has become a hot issue in the field of intelligent transportation. The automatic detection system of traffic events is the key to intelligent and automatic traffic video surveillance system. It provides conditions for dealing with traffic incidents quickly, reducing traffic delays and avoiding secondary traffic accidents. It provides a new breakthrough for highway operation and management. However, how to realize the automatic detection of traffic events efficiently, accurately and quickly is still a big problem in the field of intelligent transportation. Based on the practical application, this paper takes the video sequence of freeway as the research object, studies several key technologies such as moving target detection, tracking and abnormal behavior description, and designs the retrograde and parking of expressway. An algorithm for automatic detection of abnormal events with variable traces. In this paper, the research of the above three kinds of abnormal events mainly includes the following aspects: in the aspect of moving target detection, the background model is established by means method, and the foreground of moving target is extracted by background difference method; Aiming at the foreground of moving target with shadow, a method based on edge and HSV color space is proposed to remove shadow, and the whole foreground of moving target is extracted by morphological processing. It provides the foundation for effective moving target tracking. In the aspect of moving target tracking, the center of mass and area of the vehicle are taken as the basic features to track the vehicle, and the Kalman filter algorithm is used to find the optimal estimation of the moving target feature. Using Euclidean distance to calculate the difference of position distance and area size of moving target, the best match is found to complete the tracking of moving target, and the phenomenon that the tracking object will be lost due to the occlusion between vehicles. In this paper, an area screening method is proposed to track the vehicle in different ways, and finally to obtain the moving track of the vehicle, which provides the basis for the judgement of abnormal events. In the aspect of abnormal event detection, by analyzing the moving track of the vehicle, the moving direction of the vehicle can be intuitively understood, and the moving direction of the vehicle can be compared with the positive direction of the road to judge the retrograde event of the vehicle. Traffic parameters such as instantaneous velocity, acceleration, change of center of mass position can be obtained indirectly by analyzing the moving track of vehicle, and the change of these traffic parameters can be used to judge whether the vehicle has illegal parking event or not. Based on the analysis of the dispersion of the distance between the vehicle track and the reference lane, whether the vehicle has changed track event or not is judged. In this paper, the actual traffic video sequences of different sections of highway are tested, and the experimental results show that the algorithm of automatic detection of abnormal events in this paper is effective, and can accurately detect the abnormal events of retrograde, parking and changing roads. It has good practicability.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491;TP391.41

【参考文献】