摘要
利用数字图像技术,针对车辆检测与跟踪的3个关键环节提出了新的方法.在背景估计环节,验证像素亮度值的高斯分布特性,并据此提出背景自回归估计算法,该算法能同时适应白天和夜间两种光环境.在多个运动对象的检测环节,提出并论证一种新的只需遍历像素1次的连通成分标记算法.在车辆跟踪环节,采用Kalman滤波方法,给出状态转移矩阵和观测矩阵,并讨论初始矢量的获取模型.另外,为解决车辆跟踪过程中常出现的半遮挡问题,利用图像相似度来匹配局部图块和全图块.实际道路上的实验表明,所提出的方法实用有效,其中车辆跟踪的准确率超过95%.
This paper proposes some new methods for the three key steps of vehicle detection and trajectory tracking based on the digital image processing. In the background estimation, the Gaussian distribution hypothesis is verified and an autoregression background estimation algorithm is presented for both daytime and nighttime light-environments. In the detection of multiple moving objects, a new traversed labeled algorithm is proposed and verified,which traverses the pixels for only one time. In the tracking of vehicles, the Kalman fihering is adopted to obtain the transition and observation matrixs, and the method to get the first state vector of Kalman filter is also studied.Moreover, the image similarity is used to match the partial image to the original one, thus overcoming the semishelter usually existing in the vehicle tracking. Experimental results in real traffic scene indicate that the proposed approaches are practical and effective, with a tracking accuracy of more than 95 %.
出处
《华南理工大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2008年第6期84-89,共6页
Journal of South China University of Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(50578064)
广州市科技攻关项目(2007Z2-D3111)
关键词
车辆检测
数字图像处理
背景估计
连通像素标记
轨迹跟踪
KALMAN滤波
detection of vihicle
digital image processing
background estimation
connected pixels labeling
trajectory tracking
Kalman filtering