基于虚拟场景车辆的识别与定位技术研究  

Research on Vehicle Identification and Location Technology Based on Virtual Scence

作　　者：李占旗[1] 刘全周[1] 贾鹏飞 王启配 王述勇 LI Zhan-qi;LIU Quan-zhou;JIA Peng-fei;WANG Qi-pei;WANG Shu-yong(China Automotive technology and Research Center Co., Ltd., Tianjin 300300)

机构地区：[1]中国汽车技术研究中心

出　　处：《新型工业化》2019年第7期82-88,共7页The Journal of New Industrialization

基　　金：天津市科技计划项目(17YDLJGX00020)

摘　　要：为了提高汽车辅助系统对于目标感知能力,本文利用了机器视觉技术和智能算法对目标车辆进行了定位与跟踪。借助于dSPACE仿真软件建立虚拟交通场景,配置FasterRcnn深度学习框架并制作汽车数据样本对网络进行训练,对识别的边框进行紧缩,实现了对场景中车辆的准确检测,同时采用曲线拟合求得目标距离信息,通过卡尔曼滤波算法对检测的目标车辆进行了追踪,求得车辆距离信息的最优估计值,提高了目标车辆的定位与追踪能力。In order to improve the target recognition ability of the vehicle auxiliary system, this paper uses machine vision technology and intelligent algorithms to locate and track the target vehicle. The dSPACE simulation software is used to establish the virtual traffic scene. And the Faster Rcnn deep learning framework is configured and the car data samples are produced to train the network. The border of the recognition is corrected to realize the accurate identification of the vehicles in the scene. The target distance information is obtained by curve fitting. The image information data and the target distance information are merged by Kalman filter algorithm to obtain the optimal estimation value of image position and vehicle distance, which improves the target vehicle's tracking ability.

关 键 词：目标感知 深度学习 卡尔曼滤波 最优估计 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]