面向视频的目标跟踪新算法及其应用研究

项目名称： 面向视频的目标跟踪新算法及其应用研究

项目编号： No.61070121

项目类型： 面上项目

立项/批准年度： 2011

项目学科： 金属学与金属工艺

项目作者： 王洪元

作者单位： 常州大学

项目金额： 11万元

中文摘要： 本课题主要研究面向视频的目标跟踪的新理论与新方法。近年来，随着计算机技术的发展，面向视频的目标跟踪问题逐渐成为机器视觉领域的研究热点。本课题的研究内容主要包括以下五个方面：（1）研究了经典的粒子滤波算法与均值漂移算法的理论与算法优缺点；（2）为解决粒子滤波算法的框架下常见的粒子退化问题，提出两种解决方法：基于球粒子滤波的跟踪算法和基于局部粒子密度的跟踪算法；（3）针对现有的均值漂移算法大多基于固定目标模板或简单的模板更新策略，提出特征贡献度的概念，有效摒弃背景和噪声因素干扰，使重要性特征在匹配中起到关键作用；（4）在粒子滤波框架下，基于流形上的距离度量，提出一种基于增量半监督判别分析的跟踪方法框架；（5）对于现有方法难以在复杂环境下有效检测出运动目标的问题，提出融合光流的分通道帧差目标检测方法。 本课题的研究不仅对提高图像分析和跟踪水平有重要意义，而且课题研究中提出的一些新思想和新方法无疑将丰富模式识别与计算机视觉的研究内容。

中文关键词： 特征抽取；目标跟踪；粒子滤波；均值漂移；目标检测

英文摘要： The main work of this project is the research of new theroies and methods for object tracking in video. With the fast deveploment of computer technology, object tracking in video scenes become one of the significant issues in the field of computer vision. The main points of this project are summarized as following: (1) The merits and demerits of theroies and algorithms of particle filter and mean-shift, abbr. PF and MS respectively, are studied; (2) Particle degeneration, which influences the performance of PF, is common in typical framework of PF, and we present two ways, named ball particle filter algorithm and local density of particles based PF, to solve it; (3) Considering the issue of template matching within the MS framework, we propose a novel concept called feature contribution. With the help of feature contribution, we can effectively reduce the influences caused by background and noise, and the importance feature becomes more critical; (4) Under the framework of PF and with the metric on manifold, we present a new framework based on incremental semi-supervised discriminant analysis for tracking; (5) In the view of the complex environment, the detection of moving objects can not be satisfactory using traditional methods. An object detection method based on channel differencing combined with optical flow is proposed. Research of this project plays an important role in image analysis and video tracking, and the new ideas and new methods we proposed will enrich the contents of pattern recognition and computer vision.

英文关键词： feature extraction; object tracking; Particle filter; mean shift; object detection