导　　师： 陈爱斌

学科专业： H1203

授予学位： 硕士

作　　者： ;

机构地区： 中南林业科技大学

摘　　要： 视觉目标跟踪在视频监控、图像压缩、三维重构、机器人技术等领域有着非常重要的应用。目标跟踪的难点在于物体的突然运动,目标或背景突然改变其外部表现形式,目标的非刚性结构,目标和目标之间的遮挡、目标和背景之间的遮挡,以及摄像头的运动。本文主要研究复杂背景下的运动物体的跟踪,所做的主要工作有： 1.对粒子采样和重采样的方法进行了研究,并使用粒子滤波器对非线性、非高斯系统的状态估计进行仿真实验,通过对标准粒子滤波器与粒子滤波规则化采样方法/(RPF粒子滤波/)进行对比实验,并根据实验结果对标准粒子滤波器和RPF粒子滤波器方法的跟踪误差进行了讨论。 2.对单高斯模型及混合高斯模型的理论基础、参数估计,背景更新公式进行了分析,并通过仿真实验分析了混合高斯背景模型、单高斯背景模型的处理效果的优缺点,在此基础上,为了进一步研究有新物体移入或移出当前场景中的情况,提出了改进的高斯建模算法,将风险决策引入前景目标的突变判断,并通过实验验证了该算法的有效性。 3.对传统粒子滤波预测方法进行了改进,分别计算目标颜色模型和目标纹理模型与粒子的Bhattacharyya距离,并以此作为粒子权值更新的重要依据。将基于颜色的权值和基于纹理的权值的加权和作为综合权值,它可以反映背景和目标的特征,从而有效适应了目标跟踪时的背景和目标变化。进而提出了一种基于多信息融合的自适应粒子滤波算法,该算法在粒子滤波的框架下引入了一种自适应观测模型,给出两个状态转移方程和三个量测方程,使粒子能根据背景不同的复杂度选择合适的量测模型和状态转移模型。同时为了进一步减小目标本身的形变以及光照等外界因素对目标跟踪效果的影响,在跟踪过程中自适应更新颜色直方图和纹理直方� Visual target tracking has a very important application in the field of video surveillance, image compression,3-D reconstruction, robot techno-logy and so on. The difficulty of target tracking lies in the sudden movements, or the sudden change of the external performance form of the targets or the background, non-rigid structure of the targets, the block among goals, the block between the objects and the background, and cameras movement. This paper mainly studies the moving object tracking in the complex background. The major work performs are followings: Firstly, particle sampling and re-sampling methods are studied; Simulation experiments are done according to the state estimates of nonlinear, non-Gaussian system by using particle filter; Comparative experiments are given about the standard particle filter and particle filter rules with the sampling method /(RPF particle filter/); The tracking errors of standard particle filter and RPF particle filter method are discussed according to the experimental results. Secondly, theoretical foundation of single Gaussian model and Gaussian mixture model, parameter estimation and the background update formulas are analyzed; The advantages and disadvantages of the effects of the Gaussian mixture background model and single Gaussian background model are given by simulation analysis. On this basis, in order to research the situation when there are new objects into or out of the current scene, the improved Gaussian model algorithm is proposed, which will bring risk decision into the sudden judgment of goals of the prospect. And the experiments verify the effectiveness of the proposed algorithm. The third, the traditional particle filter forecasting method is improved. It needs to calculate the two Bhattacharyya distances between the particles and target color model, and between the particles and the target texture model. The distances serve as the important basis for updating particle weights. The weighted sum of weights based on the color and texture is regar

关 键 词： 目标跟踪 目标检测 背景建模 自适应粒子滤波 距离

领　　域： [电子电信]