基于CenterNet编码优化的车辆目标检测模型  

Vehicle target detection model based on CenterNet coding optimization

作　　者：邢雪[1] 王彬 王馨田 XING Xue;WANG Bin;WANG Xintian(School of Infomation and Control Engineering,Jilin Institute of Chemical Technology,Jilin 132000,China;School of Infomation Engineering,Minzu University of China,Beijing 100074,China)

机构地区：[1]吉林化工学院信息与控制工程学院,吉林吉林132000 [2]中央民族大学信息工程学院,北京100074

出　　处：《长江信息通信》2024年第5期73-77,共5页Changjiang Information & Communications

基　　金：吉林省教育厅科学技术研究项目(No.JJKH20230306CY)。

摘　　要：为解决车辆识别系统中车辆类型识别率低的问题,文章提出了一种优化CenterNet编码的车辆目标检测模型。本模型为保证实时性采用ResNet18作为基础网络,针对原模型中检测与分类同时进行的目标检测可能会导致车辆目标信息的混淆和互相干扰的问题,本模型将检测与分类进行解耦,着重关注车辆固有特征和车型间的特有特征,有效提高分类的准确率;针对特殊车型识别不准确的问题,模型将车辆目标的宽高,形状等特征信息加入分类中进行特征学习,提高特殊车型分类的准确性。实验结果表明:在UA-DETRAC数据集上,本模型的平均精度均值提升了0.5个百分点,F1-Score提升了1.1个百分点。In order to solve the problem of low vehicle type recognition rate in vehicle recogni-tion systems,this paper proposes a vchicle targct detcction model that optimizes CenterNct cnco-ding.This model uses ResNet18 as the basic network to ensure real-time performance.In view of the problem that simultaneous target detection and classification in the original model may lead to confusion and mutual interference of vehicle target information,this model decouples detection and classification,focusing on the inherent characteristics of vehicles and the unique characteristics between models can effectively improve the accuracy of classification.In order to solve the problem of inaccurate identification of special vehicle models,the model adds the width,height,shape and other characteristic information of the vehicle target into the classifica-tion for feature learning to improve the accuracy of special vehicle classification.Experimental results show that on the UA-DETRAC data set,the average accuracy of this model has increased by 0.5 percentagc points,and the F1-Score has increcased by 1.1 percentage points.

关 键 词：车辆目标检测 编码 CenterNet 特征融合 深度学习 

分 类 号：U491.1[交通运输工程—交通运输规划与管理]