基于监控视频的车型识别技术研究

发布时间：2018-04-13 17:20

本文选题：车型识别 + 组合特征　；参考：《浙江大学》2014年硕士论文

【摘要】：车型识别作为ITS中的一个重要分支,在打击盗窃车辆、规范交通秩序、高速公路自动收费等方面具有广阔的应用前景。和其他方法相比,基于监控视频的车型识别方法不需要在目标区域另外安装其它感应设备,从而有效的降低了系统的资本投入,并且监控视频所能提供的车型信息更为丰富,故而近些年来基于监控视频的车型识别算法取得了诸多研究成果。本文在研究运动车辆检测、车型特征提取、模式识别理论的基础上,设计了一套基于车型组合特征和改进粒子群参数优化的支持向量机的车型识别系统,主要成果包括： 1.在利用基于混合高斯背景建模的背景差分法得到运动车辆的前景图像之后,结合YCrCb颜色特征和LBP纹理特征来进行阴影去除： 2.在特征选择方面,提出了利用车辆图像的组合特征,包括Hu矩特征、LBPS特征、长宽比特征来表述车辆的特征信息,可以有效的解决单一特征容易受到光照、天气、噪声等影响,以及在识别中精度有限的问题； 3.提出了一种新的LBPS特征,在LBP特征的基础上做了改进,采用分块和信息熵的思想在车辆的纹理信息中加入了空间信息,同时降低了纹理特征的维度； 4.提出了一种改进的粒子群算法来进行支持向量机的参数寻优。在速度更新公式中增加对任一随机粒子的跟随,之后引入动量项的概念,同时考虑前一个时刻和前前时刻的速度变化情况,可以有效的解决基本粒子群算法容易陷入局部极值点和后期震荡严重收敛放缓的问题。
[Abstract]:As an important branch of ITS, vehicle recognition has a broad application prospect in cracking down on theft of vehicles, standardizing traffic order and automatic toll collection on highways.Compared with other methods, the method of vehicle identification based on surveillance video does not need to install other sensing equipment in the target area, thus effectively reducing the capital investment of the system, and the monitoring video can provide more information about the vehicle type.Therefore, in recent years, vehicle recognition algorithm based on surveillance video has made a lot of research results.Based on the research of moving vehicle detection, vehicle feature extraction and pattern recognition theory, a vehicle recognition system based on vehicle combination feature and improved particle swarm optimization is designed in this paper. The main results are as follows:1.After the background difference method based on mixed Gao Si background modeling is used to get the foreground image of moving vehicle, the shadow is removed by combining YCrCb color feature and LBP texture feature.2.In the aspect of feature selection, it is put forward that the combination features of vehicle image, including Hu moment feature, LBPS feature and aspect ratio feature, can be used to express the feature information of vehicle, which can effectively solve the problem that single feature is easily affected by illumination, weather, noise and so on.And the problem of limited precision in recognition;3.A new LBPS feature is proposed, which is improved on the basis of LBP feature. The idea of partition and information entropy is used to add spatial information to the texture information of vehicle, and the dimension of texture feature is reduced.4.An improved particle swarm optimization algorithm is proposed for parameter optimization of support vector machines.The following of any random particle is added to the velocity update formula, then the concept of momentum term is introduced, and the velocity variation of the previous moment and the preceding moment is considered.It can effectively solve the problem that the basic particle swarm optimization algorithm is prone to slow down the convergence of local extremum and late oscillation.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495;TN948.6

【参考文献】