基于分布式视频网络的交叉口车辆精确定位方法

Accurate vehicle location method at an intersection based on distributed video networks

为了对交叉口车辆的位置进行准确定位,提出了一种分布式视频网络架构下车辆精确定位方法。在分布式视频网络中每处摄像机架设位置均设有2类摄像机:近景摄像机和远景摄像机。首先在近景摄像机拍摄范围内,对感兴趣区域内车辆进行身份识别,根据车牌照平面与道路平面垂直的约束条件,建立车牌照模型来对车辆精确定位;接着在远景摄像机拍摄范围内,采用融合局部二值模式(LBP)纹理特征的金字塔稀疏光流法实时跟踪车辆上局部特征点,根据特征点运动趋势相似性获得稳态特征点,来对车辆位置估计;最后根据不同摄像机检测结果,采用加权一致性信息融合算法来提高车辆定位精度。实验结果表明:该方法能对交叉口车辆位置进行精确定位。

A robust framework is given for precise vehicle localization in intersections using distributed video networks.Each intersection is equipped with short-range and long-range cameras in a distributed video network.If the vehicle is in the shooting range of the short-range camera, within the region of interest for vehicle identification,and the license plate is perpendicular to the road plane,a vehicle license plate model is used to accurately locate the vehicle position.If the vehicle is in the shooting range of the long-range camera,apyramid sparse optical flow algorithm with LBP texture features is used in real-time to track the local feature points on the vehicle to estimate the vehicle position based on stable feature points obtained from the similar motions.Finally,information is exchanged between the cameras,a weighted consensus information fusion algorithm is used to obtain a globally optimal estimate of the vehicle position.Tests show that this method can accurately locate the vehicle position at intersections.