基于粒子滤波的视频目标跟踪算法研究.doc

PAGE \* MERGEFORMATI 基于粒子滤波的视频目标 跟踪算法研究 摘 要 目标跟踪是计算机视觉领域中的一个重要研究方向，粒子滤波算法在非线性系统、非高斯噪声情况下仍能保持良好的跟踪性能，因此被应用于视频目标跟踪领域。而在现实环境中，视频图像往往存在较大的噪声和干扰，复杂环境中的视频目标跟踪算法尤其重要。 本文在介绍了贝叶斯原理和后验概率推导公式的基础上；重点分析了基于颜色特征的粒子滤波算法，选用颜色作为目标的特征，给出了基于粒子滤波的游泳者目标跟踪的具体实现步骤及实验结果分析；提出了基于粒子滤波的目标多部位跟踪的新方法，实验结果表明目标多部位跟踪算法可以提高复杂环境下的跟踪精度。 为了使目标跟踪算法的研究更具有可视化、易扩展的特点，本课题又设计了视频跟踪系统界面，将实验的结果展现在MATLAB GUI界面中。该界面具有系统主要参数人机交换功能，清晰地显示出目标跟踪的过程，并具有暂停、停止等操作，能够很好的适应视频目标跟踪系统的研究与开发。 关键词：粒子滤波算法；目标多部位跟踪；MATLAB界面 Abstract Video object detecting is an important direction in computer vision field. In the non-liner, non-Gaussian systems, particle filter algorithm can still maintain good and stable performance characteristics, so it can be applied to the field of video tracking. But in the real world, there are much noise and interference existing in the image of the video, so it is extremely important for the video target detecting algorithm in the complex environment. Firstly, the thesis narrates the Bayesian theory, which provides us with the posterior probability; on the basis of this, it focuses on analyzing the particle filter algorithm, based on color feather template. This method chooses the color information as the feature of the object, and then shows the every procedure that composes the particle filter algorithm in detail, as well as the analysis of the experiment results. Additionally, the multisite object tracking method is proposed, and lastly, the result shows that the detecting accuracy of the multisite one is much higher in the complex circumstances. In order to enhance the visualization and extension of the system, the interface (the MATLAB GUI) of the video detecting system is designed, which is to show the result of the experiment. This interface has the function of human-computer interaction with parameters that can be adjusted by users, and the function of pause and stop which can be set as necessary. This interface can be adapted to the detecting system of the video target. Key words: particle