基于视频的车流量检测

发布时间：2018-01-06 10:34

本文关键词：基于视频的车流量检测　出处：《兰州理工大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着国民经济的快速发展,汽车的保有量与日俱增,各种各样的交通问题也随之而来,因此建立一种能实现交通信息实时检测、共享、交流的智能交通系统(ITS)就显得尤为重要。作为ITS的组成部分,基于视频的车流量检测技术具有信息量丰富、设置灵活、成本低等优点。本文对基于视频的车流量检测统计技术中的检测与统计算法进行了研究,其主要内容包括以下几个方面:(1)检测区域的设置及几何校正。首先对视频数据进行采集及预处理,然后在视频数据的首帧中手动设置车道、以确定检测区域,并用几何变换对目标图像进行校正。(2)对运动目标检测算法进行了分析、改进。对比分析了传统的目标检测方法,针对实际交通场景的视频特性,将帧间差分法和背景差分法相结合实现运动目标的检测,同时通过背景建模的方式对背景进行实时更新,并对这几个的算法做了改进,使得检测目标更加完整可靠和准确。(3)对目标特征提取算法进行了分析及改进。从检测区域二值化图像中提取车辆信息数据流,确定出运动目标区域的二维边界,然后依据目标区域车尾的中心位置选取合适的匹配准则并统计车辆数目。和传统算法相比不仅算法的执行效率高而且车辆可以实现多车道、跨车道的同时计数。仿真实验结果表明:本文的算法能够比较准确的检测到经过路口的每一辆车辆,同时也可以统计出一段时间内道路的交通流量。车流量的统计效果比较稳定,能够保持在95%以上的准确率。
[Abstract]:With the rapid development of the national economy, the amount of the automobile traffic problems grow with each passing day, then, so the establishment of a real-time detection, traffic information sharing, communication, intelligent transportation system (ITS) is particularly important. As a part of ITS, vehicle flow detection technology of video with abundant information, set up a flexible based on low cost. This paper made a research on detection and statistical algorithm of traffic flow detection in video based on statistical techniques, the main contents include the following aspects: (1) detection region setting and geometric correction. Firstly, acquisition and preprocessing of video data, and then the lane is set manually in the first frame video data, to determine the detection area and the target image is corrected by geometric transformation. (2) of the moving target detection algorithm is analyzed, the comparative analysis of improvement. Target detection system, aiming at the features of actual traffic scenes, the frame difference method and background difference method combined with the detection of moving objects, and through background modeling for real-time updates on the background, and the algorithm has been improved, so that the detection target is more complete and reliable and accurate. (3) the target feature extraction algorithm was analyzed and improved. The detection area binarization extracting vehicle information and data flow image, determine the two-dimensional boundary of target region, and then based on the center of the target area rear to select the appropriate matching criterion and count the number of vehicles. Compared with the traditional algorithm not only algorithm the implementation of high efficiency and can realize multi Lane vehicle, cross lane count. The simulation results show that this algorithm can detect accurately through the intersection to each car car, with The traffic flow can be counted for a period of time. The statistical effect of the traffic flow is stable, and the accuracy of the traffic can be kept above 95%.

【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】