基于监控视频的交通信息提取技术研究

发布时间：2018-11-06 19:41

【摘要】：当今社会,车辆迅速增长,人们对交通网络的需求也越来越强烈,只是依靠加宽道路已经不能解决问题,这样的情况下,如何对交通网络更为高效地管理便成为关键,而此时智能交通系统的出现为这一问题的解决提供了方向,对监控视频中交通信息的提取是这个领域的重点研究内容,它可以为做出交通决策提供有力支持,缓解交通压力,对交通车辆进行合理调度,使交通网络得以高效地进行运作。首先,本文对获取到的视频图像做一定的处理,使得图像达到可以实现车辆检测的要求,这些处理包括灰度的转换,直方图均的衡化,去噪,二值化,形态学操作等内容,然后仿真得出结果,并进行了对比,选出最优的处理方法。然后,再检测运动车辆,这是车辆跟踪和交通信息提取的重要前提,通过对比各种方法,并且通过对比实验结果,选取了背景差分法,因为背景差分法可以获取到完整的车辆,而且这种方法的计算量小,可以应用于对实时性要求高的场景中。之后需要对背景建模,通过改进常用的背景建模方法,使得静止或运行速度很慢的车辆不会被当作背景,能快速准确获取到背景。针对车辆阴影的存在,改进了常用的车辆阴影去除方法,这种方法相比于常用的阴影去除方法要更加有效,阴影去除完全而且不会丢失车辆信息。其次,本文实现了车辆的跟踪,这对接下来的车辆信息提取有着很关键的作用。通过对比跟踪方法,本文最终选取了Camshift算法来跟踪车辆,并引入了卡尔曼滤波器预估车辆接下来的运动情况,减小了对车辆位置搜索的范围,很大程度上加快了运算速度,使得车辆跟踪更为高效,并且由于Camshift算法的特性,车辆遮挡问题也能在一定程度上得到解决。最后,本文对交通信息进行了提取,其中包括对车速进行测量,对车流量进行获取,检测车辆是否违章停车以及车辆是否逆行,并且对这些要获取的交通信息,实验得出结果,根据结果可以得出结论,这些方法的准确性和速度都能满足要求。在车速车辆测速中,改进了常用的车辆测速方法,使得测速结果更加精确。在违章停车的检测中,改进了常用的违章停车检测方法,这种方法的检测速度快,适合在监控视频中这种对实行性要求高的场景下使用。
[Abstract]:In today's society, with the rapid growth of vehicles, people's demand for transportation network is becoming more and more intense. It is not possible to solve the problem only by widening the road. In this case, how to manage the transportation network more efficiently becomes the key. At this time, the emergence of intelligent transportation system provides a direction for the solution of this problem. The extraction of traffic information from surveillance video is the key research content in this field. It can provide strong support for making traffic decisions and relieve traffic pressure. The traffic network can be operated efficiently by reasonable dispatching of traffic vehicles. First of all, this paper does some processing to the obtained video image, which makes the image meet the requirements of vehicle detection. These processes include grayscale conversion, histogram equalization, de-noising, binarization, morphological operation, and so on. Then the simulation results are obtained and compared, and the optimal processing method is selected. Then, the moving vehicle is detected, which is an important prerequisite for vehicle tracking and traffic information extraction. By comparing various methods and comparing the experimental results, the background difference method is selected, because the background differential method can obtain the complete vehicle. Moreover, this method can be used in the scene with high real-time requirement because of its small computational complexity. After that, the background modeling is needed. By improving the common background modeling methods, the stationary or slow moving vehicles will not be regarded as the background, and the background can be obtained quickly and accurately. In view of the existence of vehicle shadow, the common methods of vehicle shadow removal are improved. This method is more effective than the usual shadow removal method, and the shadow removal is complete and does not lose vehicle information. Secondly, this paper realizes the vehicle tracking, which plays a key role in the next vehicle information extraction. By comparing the tracking method, this paper selects the Camshift algorithm to track the vehicle, and introduces the Kalman filter to estimate the next motion of the vehicle, which reduces the range of the vehicle position search, and speeds up the calculation speed to a great extent. It makes vehicle tracking more efficient, and because of the characteristics of Camshift algorithm, the problem of vehicle occlusion can be solved to some extent. Finally, this paper extracts the traffic information, including the measurement of speed, the acquisition of traffic flow, the detection of whether the vehicle stops illegally and whether the vehicle is retrograde, and the traffic information to be obtained. According to the results, it can be concluded that the accuracy and speed of these methods can meet the requirements. In vehicle speed measurement, the commonly used vehicle speed measurement method is improved to make the speed measurement more accurate. In the detection of illegal parking, the commonly used detection method of illegal parking is improved. The detection speed of this method is fast, and it is suitable for use in the scene with high performance requirements in the surveillance video.
【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】