卡尔曼粒子滤波的视频车辆跟踪算法研究  

Research on video vehicle tracking algorithm based on Kalman and particle filter

作　　者：王相海[1,2] 方玲玲[1] 丛志环[1] 

机构地区：[1]辽宁师范大学计算机与信息技术学院,大连116029 [2]南京大学计算机软件新技术国家蓐点实验室,南京210093

出　　处：《中国图象图形学报》2010年第11期1615-1622,共8页Journal of Image and Graphics

基　　金：辽宁省自然基金项目(20102123);辽宁“百千万人才工程”基金项目(2008921036);南京邮电学院图像处理与图像通信江苏省重点实验室开放基金项目(ZK207008)

摘　　要：近年来,视频车辆跟踪作为城市智能交通系统(ITS)的一个关键技术受到关注。针对传统粒子滤波的非线性、非高斯性可能导致跟踪过程的不准确性,提出一种基于卡尔曼(Kalman)粒子滤波的视频车辆跟踪算法,该算法利用基于重要区域的目标颜色直方图统计模型对视频车辆目标进行建模,并将其应用于Kalman滤波更新中,通过采用MeanShift算法将Kalman滤波器引用到粒子滤波器当中,对车辆的运行轨迹进行校正,实现了局部线性滤波,实现了在保持跟踪系统整体上的非线性、非高斯性的同时,兼顾其局部的线性高斯特性。实验结果表明,本文方法与传统粒子滤波方法相比,即使在复杂的环境下,也能够较准确地对车辆进行跟踪。Recently, video vehicle tracking as a key technology of intelligent transportation system（ITS） has got more attention. This paper introduces a video vehicle tracking algorithm based on Kalman and particle filter. The algorithm improves the traditional particle filter, whose non-linear and non-Gaussian may result in non-robustness of tracking process, the algorithm uses the targets color histogram statistical model based on the key regional to model video vehicle, and applies it to update Kalman filter. Then through the use of Mean Shift algorithm, the Kalman filter is added to the particle filter to calibrated the vehicle running tracking so that the experiment achieves a partial linear filtering, maintaining tracking system as a whole on the non-linear and non-Gaussian, and at the same time it takes into account the local characteristics of a linear Gaussian. Experimental results show that the proposed method in comparison with the traditional particle filtering can be more accurate on tracking of vehicles and ensure the robustness of performance in a complex environment.

关 键 词：视频车辆跟踪 粒子滤波 卡尔曼滤波 Mean SHIFT 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]