基于采样理论的序列蒙特卡罗车辆跟踪算法  

Sample-based Sequential Monte Carlo in Vehicle Tracking

作　　者：毛燕芬[1] 施鹏飞[1] 

机构地区：[1]上海交通大学图象处理与模式识别研究所,上海200030

出　　处：《系统仿真学报》2004年第11期2520-2521,2528,共3页Journal of System Simulation

基　　金：国家"973"项目(No.TG1998030408)

摘　　要：结合图像的灰度分布和空间布局,提出了一种基于灰度和空间信息的序列蒙特卡罗概率跟踪算法。通过比较采样值和期望值的特征距离来计算采样状态对应的权值。利用加权采样值来估计未知量后验概率的各阶矩,当采样数趋向于无穷时由大数定理保证,采样值分布逼近于真实值分布。仿真实验给出了利用因子采样对交通流中车辆跟踪的结果。实验表明,基于序列蒙特卡罗的车辆跟踪方法计算简单有效,可以准确地得到车辆的位置并且很好地跟踪其运动轨迹。Incorporating intensity distribution and spatial layout, this paper proposes a sequential Monte Carlo probabilistic tracking algorithm using intensity and spatial information. By computing the distance between sample and target, different weights associated with every sample and the moments of the posterior state vector can be computed. The samples distribution trends to the state distribution, whose validity is guaranteed by the strong law of large numbers (SLLN). Experimental results show the efficient vehicle tracking performance using factored sampling in traffic scene.

关 键 词：贝叶斯估计 序列蒙特卡罗滤波 因子采样 概率跟踪 仿真 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]