摘要
车速和车型作为重要的车辆信息,在道路监控系统中发挥着很大的作用.传统的基于视觉的车辆信息识别方式由于计算参数过大且提取的特征不足,难以满足智能交通实时性和普适性的需求.对此,提出了一种新的车辆信息识别方法,采用运动目标检测技术实现视频中车辆的提取,然后利用虚拟线圈法进行车速识别,再通过改进的残差网络对提取的车辆进行车型识别,有效地减少了计算参数,实现了对视频的快速处理,同时利用了残差网络极强的特征表达能力,提高了识别的准确率.此外,加入了重载车型的研究,有良好的应用前景.实验结果显示,系统车速识别平均绝对误差不超过6km/h,车型识别平均准确率达到92.1%,针对小客车和小轿车的识别准确率高达98.7%,优于传统的识别方法.
As the important vehicle information,speed and type of vehicle play an important role in the road monitoring system.The traditional vision-based identification of vehicle is weak in real-time performance and universality,because the calculation parameters are too large and the abstraction of extracted feature is insufficient.In this regard,a new vehicle information recognition method is proposed.The vehicle extraction is realized by moving target detection,then the vehicle speed is identified by the virtual coil method,and the vehicle type is classified by the improved residual network.This method effectively reduces the calculation parameters,realizes the rapid processing of video,and utilizes the strong feature expression ability of the residual network to improve the accuracy of recognition.In addition,the classification of heavy vehicles has been added,which provides a good application prospects.Experimental results show that the average absolute error of speed identification is not more than 6 km/h,and the average precision of vehicle type identification is 92.1%,for minibuses and cars,the precision is as high as 98.7%,which is superior to the traditional identification method.
作者
梁栋
何佳
石陆魁
王松
刘佳
LIANG Dong;HE Jia;SHI Lukui;WANG Song;LIU Jia(School of Civil and Transportation Engineering,Hebei University of Technology,Tianjin 300000,China;School of Artificial Intelligence,Hebei University of Technology,Tianjin 300000,China)
出处
《北京交通大学学报》
CAS
CSCD
北大核心
2019年第5期10-19,共10页
JOURNAL OF BEIJING JIAOTONG UNIVERSITY
基金
国家自然科学基金(50808063)~~
关键词
智能交通系统
卷积神经网络
车型
车速
智能识别
intelligent transportation system
convolutional neural network
vehicle type
vehicle speed
intelligent identification
