基于视频的夜间车流量检测研究

发布时间：2018-02-08 12:59

本文关键词： 车流量检测夜间交通监控运动信息车灯侦测配对高检测率　出处：《西北师范大学》2015年硕士论文　论文类型：学位论文

【摘要】：车辆检测是智能交通系统(intelligent transportation system,ITS)的重要组成部分,因此基于视频的车辆检测器成为智能交通系统的研究热点并得到越来越广泛的应用。视频车辆检测系统能很好的获取交通参数,且安装和检修便利,具有事故侦测和视频监控等优点。白天视频检测技术已经有大量学者做了很多工作,且检测技术已经发展成熟,能满足城市智能交通系统的需要,然而夜间照明度低和光线反射是视频车辆检测器实现全天候工作的最大难题。传统的夜间车辆检测主要从两个方面来克服夜间光线不足,各种光线干扰的影响:一是检测夜间车辆最明显的特征——车尾灯,该方法的鲁棒性不强,易受后面车辆车灯照射和车身边界反光的影响,交通信息的提取有限,对摄像头的装置角度和镜头区域有要求,且昼夜模式之间的转换是个难题;二是从摄像机本身出发,选择高性能具有红外捕捉功能的摄像机或是提高侦测区域照明条件获得车辆信息,该方法加大了系统开销,且系统的安装和维护难度加大。车辆本身在夜间视频图像中可识别信息量少,光照强度变化大,再加上路面各种光源的反射干扰,使得车型不易判断,导致运动目标提取困难,误检率也较高。为了更好地解决上述难题,本文提出一种包含帧间差和背景差结合背景更新的车灯配对跟踪的夜间车辆检测方法。首先,采用前述算法获取高亮度车灯区域,以消除夜晚光照强度急剧变化造成的光晕。其次利用车灯亮度、颜色、形状、位置、纹理等运动信息找出可能为车灯的亮点,然后再判断属于同一车的可能性,并进行车灯配对,最后设置虚拟检测线,针对路面上碰触虚拟检测线的一对亮点进行计数。实验结果表明:该算法简洁,帧处理速度快,具有较高的实时性、鲁棒性,即使在复杂天气(雨天,雪天,雾霾)和交通堵塞情况下,也有较高的检测率。本方法在夜间车辆检测的平均准确率在96%以上,能够满足城市道路及高速公路的实时检测需求。
[Abstract]:Vehicle detection is an important part of intelligent transportation system. Therefore, the vehicle detector based on video has become the research hotspot of the intelligent transportation system and has been more and more widely used. The video vehicle detection system can obtain the traffic parameters well, and the installation and maintenance are convenient. Many scholars have done a lot of work in the daytime video detection technology, and the detection technology has been developed to meet the needs of the urban intelligent transportation system. However, low night illumination and light reflection are the most difficult problems for video vehicle detectors to achieve all-weather operation. Traditional night vehicle detection mainly overcomes the lack of light at night from two aspects. The first is the detection of the most obvious feature of the vehicle at night, that is, the vehicle taillights. This method is not robust enough, and is vulnerable to the influence of the rear vehicle lights and the reflection of the body boundary, so the extraction of traffic information is limited. There are requirements for the camera's device angle and lens area, and the conversion between day and night modes is a problem; second, starting from the camera itself, Choosing a camera with high performance and infrared capture function or improving the illumination condition of detecting area to obtain vehicle information, this method increases the cost of the system. It is difficult to install and maintain the system. The vehicle itself has less identifiable information in the video images at night, the light intensity changes greatly, and the reflection interference of all kinds of light sources on the road makes it difficult to judge the vehicle, which leads to the difficulty of extracting moving targets. In order to solve the above problem better, this paper proposes a new method of vehicle detection, which includes inter-frame difference and background difference, combined with background updating. Firstly, the algorithm is used to obtain the region of high-brightness vehicle lamp. In order to eliminate the halo caused by the sharp change of light intensity at night. Secondly, using the light intensity, color, shape, position, texture and other moving information to find out the possible bright spot of the lamp, and then determine the possibility of belonging to the same car. Finally, the virtual detection line is set up, and a pair of bright spots of the virtual detection line are counted. The experimental results show that the algorithm is simple, the frame processing speed is fast, the real-time performance is high, and the robustness is high. Even in complicated weather (rainy, snowy, haze) and traffic jam, there is a high detection rate. The average accuracy of this method in night vehicle detection is more than 96%, which can meet the needs of real-time detection of urban roads and highways.
【学位授予单位】：西北师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495

【相似文献】