基于特征融合和度量学习的车辆重识别  

Vehicle Re-identification Based on Multi-features and Metric Learning

作　　者：王盼盼 李玉惠[1] 李福卫 WANG Panpan;LI Yuhui;LI Fuwei(School of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China;Juxinfeng Science and Technology Co.,Ltd.of Kunming,Kunming 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明云南650500 [2]昆明聚信丰科技有限公司,昆明云南650500

出　　处：《电子科技》2018年第9期29-31,79,共4页Electronic Science and Technology

基　　金：国家自然科学基金(61363043);云南省创新资金项目(2016EH076)

摘　　要：在不同的道路交通视频监控拍摄条件下,识别出同一车辆是车辆重识别需要解决的主要问题。针对车辆重识别时不同拍摄视角、光照等拍摄条件同一车辆的视频监控图像存在差异的问题,提出一种结合特征融合和度量学习的车辆重识别方法。利用Local Maximal Occurrence(LOMO)方法对车辆样本进行特征表示,该特征提取方法可以有效的降低外界拍摄条件对识别率的影响,对提取的特征数据进行LDA降维可减少计算复杂度提高分类精度,并通过马氏距离(Mahalanobis distance)对车辆样本进行精确的重识别。实验结果表明,该方法在车辆重识别方面具有较高的识别率,且对光照变化、视角变化都具有较好的鲁棒性。Under the condition of different road traffic video surveillance shooting, it is recognized that the same vehicle is the main problem that the vehicle re-identification needs to be solved. Aiming at the problem that the video surveillance images of the same vehicle are different in different shooting angles and illumination conditions, a method of vehicle re - recognition based on feature fusion and metric learning is proposed. Firstly, the Local Maximal Occurrence （LOMO） method is used to express the characteristics of the vehicle samples. The feature extraction method can effectively reduce the influence of the external shooting conditions on the recognition rate. LDA reduction of the extracted feature data can reduce the computational complexity and improve the classification accuracy , And the vehicle samples are accurately re-identified by Mahalanobis distance. The experimental results show that this method has a high recognition rate in vehicle re - recognition and has good robustness to both changes in illumination and viewing angle.

关 键 词：特征融合 车辆重识别 LOMO算法 马氏距离 度量学习 

分 类 号：TN919.81[电子电信—通信与信息系统]