基于机器视觉的交通异常事件检测算法研究

发布时间：2018-05-31 16:15

本文选题：智能交通 + 交通事故检测　；参考：《电子科技大学》2017年硕士论文

【摘要】：交通事故检测是智能交通系统中最重要的部分之一,实时且鲁棒的交通事故检测方法可以在减少人员伤亡和减少财产损失上做出巨大贡献。随着智能交通系统的快速发展,基于计算机视觉技术和图像处理技术的交通事故检测系统的研究吸引了广泛的注意力,许多研究者在这个领域中也取得了重大进展。然而,由于交通环境的复杂性,目前提出的一些方法在实际应用中仍然存在一定的限制。为了保证对交通事故快速和精确地检测,并且达到实际运用的目的,一些挑战性的问题就需要被解决。这样的一种交通事故检测算法必须达到三个要求。第一,能够处理复杂的交通环境,包括较差的天气、不同的光照条件、复杂的道路状况和不同的交通参与者;第二,能够处理不同的交通流状态,也就是不同程度的交通拥堵;第三,能够实时性运行。在本文中,我们研究在交通视频流中实时和鲁棒的交通事故检测算法。传统的方法要么就是运行速度不够快,要么就是在复杂的交通环境下鲁棒性不够好。我们提出了一种新颖的基于监控视频的交通事故检测方法,该方法的主要观察点在于交通事故的发生不仅会引起局部目标的运动方向混乱,也会造成全局交通流的混乱。在我们的方法中,一共有三步。第一,对每一帧视频构建对应的光流场;第二,基于光流场构建全局交通流描述子高斯模型和局部运动方向图高斯模型,来对交通事故的全局特征和局部特征进行检测,从而对视频帧中发生的交通事故进行检测与定位;第三,构建了一个检验模块来进一步验证交通事故的发生,并排除事故误报的情况。我们的方法具有实时性运行、高精度、漏报少、误报率低以及对不同交通环境和光照条件具有鲁棒性等优点,后期通过广泛的量化评估实验验证了以上优点,并且展示了在这个领域中的巨大进步。
[Abstract]:Traffic accident detection is one of the most important parts in intelligent transportation system. The real-time and robust traffic accident detection method can make a great contribution to reducing casualties and property losses. With the rapid development of intelligent transportation system, the research of traffic accident detection system based on computer vision and image processing technology has attracted extensive attention, and many researchers have also made great progress in this field. However, due to the complexity of traffic environment, there are still some limitations in the practical application of some proposed methods. In order to ensure the rapid and accurate detection of traffic accidents and to achieve the purpose of practical application, some challenging problems need to be solved. Such a traffic accident detection algorithm must meet three requirements. Firstly, it can deal with complicated traffic environment, including bad weather, different light conditions, complex road conditions and different traffic participants; second, it can deal with different traffic flow states, that is, traffic congestion of different degrees. Third, it can run in real time. In this paper, we study real-time and robust traffic accident detection algorithms in traffic video streams. The traditional approach is either not fast enough or robust enough in complex traffic environments. We propose a novel method for traffic accident detection based on surveillance video. The main observation point of this method is that the occurrence of traffic accident will not only cause confusion of the moving direction of local target, but also cause chaos of global traffic flow. There are three steps in our method. Firstly, the corresponding optical flow field is constructed for each frame of video; secondly, the global traffic flow descriptor Gao Si model and the local motion pattern Gao Si model are constructed based on the optical flow field to detect the global and local characteristics of traffic accidents. In order to detect and locate the traffic accident in the video frame, a verification module is constructed to further verify the occurrence of the traffic accident and eliminate the false alarm of the accident. Our method has the advantages of real-time operation, high precision, low false alarm rate, low false alarm rate and robustness to different traffic environment and illumination conditions. And showed great progress in this field.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495;TP391.41

【参考文献】