基于视频的车流量检测系统研究与实现

发布时间：2018-03-06 23:30

本文选题：智能交通系统　切入点：车辆目标检测　出处：《南昌航空大学》2015年硕士论文　论文类型：学位论文

【摘要】：近年来,随着我国机动车数目大量增长所带来的困恼,如何改善交通现状成为亟待解决的社会核心问题。在此背景下,智能交通系统(Intelligent Transportation System,ITS)便应运而生。研究ITS的意义在于保障人民的出行安全,提高交通系统的工作效率,改善生活环境以及节约社会资源。而基于视频的车流量检测的研究是智能交通系统的核心内容,虽然已经取得了一些重大研究成果,但仍然存在着一些棘手的问题需要解决和完善。例如在雾天的情况下,噪声存在于整个图像序列中,对车辆目标检测具有严重的干扰;当车辆图像存在阴影时,由于车辆阴影和车辆本身具有相同的速度和运动方向,往往在检测的过程中会把阴影划归为车辆目标当中。以上情况都是客观存在和急需解决的问题。本文主要针对上述两种情况进行探讨和研究,主要内容和创新点如下:(1)针对传统车辆目标检测方法在正常环境和雾天环境下检测精度较低的问题,提出了一种基于低秩矩阵的车流量检测方法。首先利用伊辛模型和鲁棒性主成份分析方法得到非凸能量函数,随后利用奇异值分解分步解决能量函数非凸性的问题,进而通过优化能量函数来检测出理想车辆目标。实验结果表明,该方法与帧差法和混合高斯模型法相比,检测车流量的精度得到了显著提高,并且在雾天环境下也能很好的分割出车辆目标。(2)针对传统车辆目标检测方法在阴影场景中检测精度较低的问题,提出了一种融合HSV颜色空间的大区域纹理特征阴影检测方法。该方法从阴影与背景具有相似的纹理特征性出发,首先利用HSV颜色空间分离得到的强度特征信息和纯度特征信息来搜索候选阴影像素,然后根据像素关联性将候选阴影像素组合成候选阴影大区域,最后根据极坐标得到像素幅值和像素梯度方向,把具有大幅值的边界候选阴影像素作为关键像素,通过计算关键像素梯度方向关联性得到最终的阴影区域。实验结果表明,该方法可以很好的处理车辆检测中存在的阴影问题,并且具有较好的鲁棒性。
[Abstract]:In recent years, with the great increase in the number of motor vehicles in China, how to improve the traffic situation has become the core social problem to be solved. Intelligent Transportation system emerges as the times require. The significance of studying ITS is to ensure people's travel safety and improve the working efficiency of transportation system. Improving the living environment and saving social resources. And the research of vehicle flow detection based on video is the core content of intelligent transportation system, although some important research results have been achieved. However, there are still some thorny problems that need to be solved and perfected. For example, in the case of fog, noise exists in the whole image sequence, which has serious interference on vehicle target detection. Since the shadow of the vehicle and the vehicle itself have the same speed and direction of motion, In the process of detection, the shadow is often classified as the vehicle target. The above situations are both objective and urgent problems. This paper mainly discusses and studies the above two situations. The main contents and innovations are as follows: 1) aiming at the problem of low accuracy of traditional vehicle target detection methods in normal and foggy environments, In this paper, a method of vehicle flow detection based on low rank matrix is proposed. Firstly, the non-convex energy function is obtained by using Ising model and robust principal component analysis method, and then the problem of nonconvexity of energy function is solved step by step by singular value decomposition (SVD). The experimental results show that compared with the frame difference method and the mixed Gao Si model method, the accuracy of the vehicle flow detection is improved significantly. And in the fog environment can also be a good segmentation of vehicle targets. 2) aiming at the traditional vehicle target detection method in shadow scene detection accuracy is low. A shadow detection method for large area texture features based on HSV color space is proposed, which is based on the similar texture features of shadow and background. Firstly, the intensity feature information and purity feature information obtained from HSV color space are used to search candidate shadow pixels, and then candidate shadow pixels are combined into candidate shadow regions according to pixel correlation. Finally, the pixel amplitude and pixel gradient direction are obtained according to polar coordinates, and the edge candidate shadow pixels with large values are taken as key pixels. Finally, the final shadow region is obtained by calculating the correlation of the key pixel gradient direction. The experimental results show that, This method can deal with the shadow problem in vehicle detection and has good robustness.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U491.116

【相似文献】