基于组件超分辨率的多分辨率车辆重识别方法  

Multi-Resolution Vehicle Re-Identification Method Based on Component Super-Resolution

作　　者：黄文心 钟忺[2] 张军[2] 巫世峰 陈淑琴 李琳[2] 刘文璇 HUANG Wen-xin;ZHONG Xian;ZHANG Jun;WU Shi-feng;CHEN Shu-qin;LI Lin;LIU Wen-xuan(School of Computer Science and Information Engineering,Hubei University,Wuhan 430062,China;School of Computer Science and Artificial Intelligence,Wuhan University of Technology,Wuhan 430070,China;ZhongQianLiYuan Engineering Consulting Co Ltd,Wuhan 430070,China;College of Computer,Hubei University of Education,Wuhan 430205,China)

机构地区：[1]湖北大学计算机与信息工程学院,武汉430062 [2]武汉理工大学计算机与人工智能学院,武汉430070 [3]中乾立源工程咨询有限公司,武汉430070 [4]湖北第二师范学院计算机学院,武汉430205

出　　处：《武汉理工大学学报》2022年第11期96-104,共9页Journal of Wuhan University of Technology

基　　金：湖北省自然科学基金(2021CFB281,2021CFB513,2021BAA030);国家自然科学基金(62271361,62276196)

摘　　要：在真实的道路监控场景中,考虑到监控摄像头与车辆的距离、监控摄像头不同的朝向和监控摄像头的高度等因素,捕捉到的车辆图片的分辨率差异较大,严重影响车辆重识别的性能。为此,提出基于对比学习的多分辨率特征提取方法,利用组件分割器提取车辆的组件特征,并通过对比学习的方法融合多分辨率输入图像的全局与组件特征,充分提取车辆整体边缘和细节信息。同时提出基于组件超分辨率的特征增强方法,对车辆图像进行组件分割并去除背景,将超分辨率应用到车辆组件上增强组件表征,可有效避免全局噪声干扰。与近年来先进的重识别方法开展了大量对比实验,在MLR-VeRi776和VRIC数据集上开展的实验结果验证了提出的多分辨率车辆重识别方法具备一定优势。In a real road monitoring scenario,considering the distance between the surveillance camera and the vehicle,the different orientations of the surveillance camera,and the height of the surveillance camera,the resolution difference of the captured vehicle images are obvious,which seriously affects the performance of vehicle re-identification.To this end,a multi-resolution feature extraction method based on contrastive learning is proposed.The component segmenter is used to extract the component features of the vehicle,and the global and component features of the multi-resolution input image are fused by the method of contrastive learning.At the same time,a feature enhancement method based on component super-resolution is proposed to segment the vehicle image and remove the background and apply super-resolution to vehicle components to enhance the component representation,which can effectively avoid global noise interference.A large number of comparative experiments have been carried out with advanced re-identification methods in recent years.The experimental results carried out on the MLR-VeRi776and VRIC datasets verify that the proposed multi-resolution vehicle re-identification method has certain excellence.

关 键 词：多分辨率车辆重识别 组件分割 对比学习 组件超分辨率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]