道路交通环境下车辆前景提取算法研究

发布时间：2019-05-19 06:53

【摘要】：智能交通系统(Intelligent Transportation Systems,ITS)涵盖了交通领域的方方面面,基于视频图像的交通流数据采集与交通行为自动分析顺应了ITS的发展,成为ITS的热点研究领域。基于图像处理的车辆前景提取作为这一研究中基础而又重要的步骤之一,在ITS发展中,有着重要的理论研究意义和潜在的应用价值。本文将车辆前景提取分为车辆运动检测和车辆阴影消除两个步骤,以车辆前景提取为目的,对车辆运动检测和车辆阴影消除进行分析与研究。首先,详细分析混合高斯模型(Gaussian Mixture Model,GMM)原理并阐述GMM缺陷及其原因。针对GMM背景建模简单,易受初始时刻车辆前景干扰的问题,本文引入格拉布斯异常值判别准则,提出了基于时域约束平均的GMM背景建模方法;依据改进的帧间差分法和GMM背景更新原理,提出了基于四分区自适应GMM的车辆运动检测算法,以此解决传统GMM存在的车辆空洞问题、ghost问题和车辆停车误检测问题。然后,在对阴影分析的基础上,指出本文车辆阴影消除是指对行驶车辆投射阴影的消除。针对单一采用颜色特征进行阴影消除的不足,本文通过Log域差分,弱化阴影,得到肯定是车辆前景的区域,以此对基于HSV颜色空间特征的阴影消除方法进行约束,提出了基于Log域差分和HSV颜色空间特征的车辆阴影消除方法。最后,对所提算法进行仿真实验分析,并依据基本的数学知识,定义了空洞填补率,以此对车辆空洞问题的解决程度进行量化分析。实验结果表明,所提出的改进GMM算法可以克服原GMM缺陷,所提出的车辆阴影消除方法性能良好。验证了本文通过车辆运动检测和车辆阴影消除来提取车辆前景的可靠性。
[Abstract]:Intelligent Transportation system (Intelligent Transportation Systems,ITS) covers all aspects of traffic field. Traffic flow data acquisition and traffic behavior automatic analysis based on video image conform to the development of ITS and become a hot research field of ITS. As one of the basic and important steps in this research, vehicle prospect extraction based on image processing has important theoretical research significance and potential application value in the development of ITS. In this paper, vehicle foreground extraction is divided into two steps: vehicle motion detection and vehicle shadow elimination. For the purpose of vehicle foreground extraction, vehicle motion detection and vehicle shadow elimination are analyzed and studied. First of all, the principle of mixed Gao Si model (Gaussian Mixture Model,GMM is analyzed in detail, and the defects of GMM and its causes are expounded. In order to solve the problem that the background modeling of GMM is simple and vulnerable to the disturbance of vehicle prospect at the initial time, this paper introduces the Grabbs anomaly criterion, and proposes a GMM background modeling method based on time-domain constrained average. According to the improved inter-frame difference method and GMM background updating principle, a vehicle motion detection algorithm based on four-partition adaptive GMM is proposed to solve the vehicle hole problem, ghost problem and vehicle parking error detection problem existing in traditional GMM. Then, on the basis of shadow analysis, it is pointed out that the shadow elimination of the vehicle in this paper refers to the elimination of the shadow cast on the driving vehicle. In order to solve the problem of shadow elimination by using color features alone, this paper weakens the shadow by Log domain difference, and obtains the area which is sure to be the foreground of the vehicle, so as to constrain the shadow elimination method based on the color space feature of HSV. A vehicle shadow elimination method based on log domain difference and HSV color space feature is proposed. Finally, the proposed algorithm is analyzed by simulation experiments, and the filling rate of voids is defined according to the basic mathematical knowledge, so as to quantitatively analyze the degree of solving the problem of vehicle voids. The experimental results show that the improved GMM algorithm can overcome the defects of the original GMM, and the proposed vehicle shadow elimination method has good performance. The reliability of vehicle prospect extraction through vehicle motion detection and vehicle shadow elimination is verified.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495;TP391.41

【相似文献】