计算机工程与应用 ›› 2012, Vol. 48 ›› Issue (6): 24-27.

• 博士论坛 • 上一篇    下一篇

序列蒙特卡罗框架中引入对数取样的重采样

吴 刚1,2,唐振民2,杨静宇2   

  1. 1.南京工程学院 车辆工程系,南京 211167
    2.南京理工大学 计算机科学与技术学院,南京 210094
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2012-02-21 发布日期:2012-02-21

Resampling strategy imported by logarithm sampling in sequential Monte Carlo framework

WU Gang1,2, TANG Zhenmin2, YANG Jingyu2   

  1. 1.Department of Vehicle Engineering, Nanjing Institute of Technology, Nanjing 211167, China
    2.School of Computer Science and Technology, Nanjing University of Science and Technology, Nanjing 210094, China
  • Received:1900-01-01 Revised:1900-01-01 Online:2012-02-21 Published:2012-02-21

摘要: 为解决粒子退化问题,在序列蒙特卡罗理论方面提出了一种引入对数取样的重采样策略,从仿真实验角度证实了它的有效性。对比其他几种典型的重采样,提出的重采样策略系统误差最小。在后验均值误差和均方差两项主要指标上,引入对数取样后的重采样由于平滑了采样点的分布,因此降低了重采样策略的系统误差;实验将提出的重采样策略嵌入到目标跟踪算法中,实际的测试结果同样验证了该重采样的收敛性和良好的抗噪性能。该理论性方法不仅适用于计算机视觉系统,而且可以应用于广泛的属于时间序列分析的非线性非高斯系统。

关键词: 蒙特卡罗框架, 重采样, 对数取样

Abstract: Aiming at resolving degeneration of particles, an improved resampling strategy is brought forward by importing logarithm sampling on Sequential Monte Carlo theory. Its efficiency is confirmed from emulation mode. Compared with several traditional resampling strategies, the system error about the improved strategy is lowest. Due to importing logarithm sampling on particles smoothly, the errors about resampling strategy are reduced especially on the two important parameters Posterior Mean Error(PME) and Mean Square Error(MSE). The improved resampling strategy is embedded in object tracking algorithm. The experimental results show that astringency and antinoise-capability about the improved strategy are excellent in real scene. The theoretical method can be used in computer vision system, moreover it can be used in non-linear and non-Gaussian system of time series analysis.

Key words: Monte Carlo framework, resampling, logarithm sampling