一种融入PCA的LBP特征降维车型识别算法  

A model recognition algorithm integrating PCA into LBP feature dimension reduction

作　　者：董恩增[1] 魏魁祥 于晓[1] 冯倩[1] 

机构地区：[1]天津理工大学复杂系统控制理论及应用重点实验室,天津300384

出　　处：《计算机工程与科学》2017年第2期359-363,共5页Computer Engineering & Science

基　　金：国家自然科学基金(61172185);天津市高等学校科技发展基金(20120829)

摘　　要：车型识别是智能交通系统研究的关键技术之一,针对车型识别的过程中存在处理的信息量大,提取特征维数高,识别实时性较差等问题,设计了一种融入PCA的LBP特征降维车型识别算法。首先在视频序列中使用帧间差分法提取目标车辆;然后计算目标车辆的LBP特征并利用PCA方法将数据由像素维数降至训练数据维数,在增强识别算法对光线变化鲁棒性的同时,一定程度上降低了车型识别的计算量;最后利用最小距离分类器对目标车辆进行分类识别。实验结果表明,所设计的车型识别算法与常规PCA方法相比,所设计的算法在光照变化时识别准确率有所提高,算法的实时性得到了一定的提升。Vehicle recognition is one of the key technologies in the study of the intelligent transportation system. Aiming at the problems of a large amount of information to be managed, high dimension of extracted features and poor real-time recognition performance in the process of vehicle recognition, we propose a model recognition algorithm integrating PCA into LBP feature dimension reduction. Firstly, we use the inter-frame difference method to extract the target vehicle in the video sequence, then calculate the LBP characteristics of the target vehicle and utilize the PCA method to reduce the data dimension from pixel dimensions to training data dimensions. This algorithm can enhance its robustness to light variance and reduce the computation amount for vehicle recognition to some extent at the same time. Finally we use the minimum distance classifier to classify the vehicle model. Experimental results show that compared with the conventional PCA method, the proposed model recognition method has higher recog- nition accuracy when light changes and good real-time performance to a certain degree.

关 键 词：车型识别 帧间差分法 特征降维 鲁棒性 最小距离分类器 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]