基于局部感知的车辆重识别算法  

Local-aware based vehicle re-identification perception

作　　者：陈冬艳 彭锦佳 蒋广琪 付先平[1,2] 米泽田 CHEN Dong-yan;PENG Jin-jia;JIANG Guang-qi;FU Xian-ping;MI Ze-tian(School of Information Science Technology,Dalian Maritime University,Dalian 116026,China;Robotics Center Department,Peng Cheng Laboratory,Shenzhen 518033,China)

机构地区：[1]大连海事大学信息科学技术学院,辽宁大连116026 [2]鹏城实验室机器人中心部,广东深圳518033

出　　处：《计算机工程与设计》2022年第7期2048-2054,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61802043、61370142、61272368);中国博士后基金项目(3620080307);兴辽英才计划基金项目(XLYC1908007);大连市科技创新基金项目(2018J12GX037、2019J11CY001);中央高校基本科研业务费专项基金项目(3132016352、3132019203、3132020215)。

摘　　要：针对车辆重识别技术中难以通过全局外观特征准确识别不同车辆之间细微差异性的问题,提出一种基于局部感知的车辆重识别算法(local-aware based vehicle re-identification,LVR)。获取全局宏观特征以保留图像的上下文信息;利用空间变换网络的对齐模块对车辆图像进行分块,获取车辆局部细节信息;采用由粗到细的关键点检测方法获取局部关键点特征。在两个大型车辆数据集(即VeRi和VehicleID)上的评估结果表明,该算法具有较好的重识别效果。Aiming at the problem that it is difficult to accurately identify the subtle differences between different vehicles through global appearance features in vehicle re-identification technology,a local-aware based vehicle re-identification(LVR)was presented.The macro characteristics of the vehicle were learnt and the context information of the image was retained.The alignment module of the spatial transformation network was used to remove the interference of the background information,and the vehicle image was blocked to obtain local detailed information of the vehicle.The key point detection method from coarse to fine was used to obtain the local key point features.The evaluate method was evaluated on two large-scale vehicle Re-ID datasets,i.e.,VeRi and VehicleID.Experimental results show that the proposed method achieves excellent performance.

关 键 词：车辆重识别 局部分块特征 局部关键点特征 特征提取 深度学习 

分 类 号：TN911.73[电子电信—通信与信息系统]