基于改进的Faster R-CNN的车辆目标检测  

Vehicle Detection Method Based on Fast R-CNN

作　　者：单志勇[1] 宫加辉 SHAN Zhiyong;GONG Jiahui(Donghua University,Shanghai 201600)

机构地区：[1]东华大学,上海201600

出　　处：《现代计算机》2021年第20期74-79,共6页Modern Computer

摘　　要：车辆目标检测作为交通管理系统的重要组成部分具有重要的研究意义。为了解决传统车辆目标检测带来的准确率低的问题,提出了基于改进的Faster R-CNN算法的车辆目标检测。改进后的Faster R-CNN算法在原始Faster RCNN算法的基础上随机选取960×540、900×500、800×480三种尺寸的训练图片进行训练,同时对RPN(Region Proposal Network)中的候选区域比例进行了扩展,增加了1:3、3:1两种比例。改进后的目标检测的mAP达到了95.56%,比基于Faster R-CNN算法的车辆目标检测的mAP提高了0.06%。As an important part of traffic management system,vehicle target detection has important research significance.In order to solve the problem of low accuracy brought by traditional vehicle target detection,a vehicle target detection method based on improved Faster R-CNN algorithm is proposed.Based on the original Faster R-CNN algorithm,the improved Faster R-CNN algorithm randomly selects 960×540,900×500 and 800×480 training images for training,and expands the proportion of candidate regions in RPN(Region Proposal Network)by 1:3 and 3:1.The mAP of improved target detection is 95.6%,which is 0.1%higher than that of vehicle target detection based on Faster RCNN algorithm.

关 键 词：Faster R-CNN算法 目标检测 深度学习 

分 类 号：U495[交通运输工程—交通运输规划与管理] TP183[交通运输工程—道路与铁道工程] TP391.41[自动化与计算机技术—控制理论与控制工程]