稀疏性多维联合优化在线视觉跟踪方法研究

项目名称： 稀疏性多维联合优化在线视觉跟踪方法研究

项目编号： No.61572285

项目类型： 面上项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 刘军清

作者单位： 三峡大学

项目金额： 16万元

中文摘要： 基于稀疏性联合模型在线视觉跟踪技术因能直观模拟人眼视觉跟踪、增强了系统的鲁棒性而成为计算机视觉领域发展的关键问题之一。本项目主要就基于联合稀疏表示视觉跟踪面临的多特征融合、异值视觉测量去除，以及传统联合方法大多停留在乘性或加性层面，难以自然直观地识别和分类目标和背景等问题，从多维联合优化的基本问题和模型着手，利用分层粒子滤波和多任务学习，实现目标与背景多特征结构化联合稀疏表示。主要研究内容为多特征结构化联合稀疏表示方法，并将研究成果应用到长时复杂智能监控视频环境的目标跟踪。

中文关键词： 目标跟踪；目标拟合；目标分割；特征提取；视频预处理

英文摘要： Since online sparse visual tracking technology have simulated human eyes visual tracking and enhanced the robustness of tracking system, online visual tracking technology based on sparse joint model is becoming one of the key problems in the field of computer vision. Traditional joint tracking methods only focus on simple collaborative methods, such as add or multiplicative. It is difficult to identify and classify the target from background intuitively. In this project, we take some fundamental issues and model in the multi-dimension joint optimization as breakthrough point; for instance, multi features fusion and unreliable visual cues detecting etc., using hierarchical particle filter and MIL to achieve multi-dimension joint optimization including multi-features structural joint sparse representation for the object and the background. The research topics of this project mainly include: the multi-features structural joint sparse representation method. The research results of this project will be used in the long-time complex intelligent video surveillance systems.

英文关键词： Object tracking；Target fitting；Object segmentation；Feature extraction；Video preprocessing