项目名称: 基于随机有限集的视频目标跟踪算法与应用研究
项目编号: No.61305016
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 吴静静
作者单位: 江南大学
项目金额: 22万元
中文摘要: 提出基于随机有限集的视频多目标跟踪新算法及其实现方法,即基于随机集的改进粒子概率假设密度(PHD)滤波器。该算法解决了基本粒子PHD滤波器在视频跟踪中的缺陷,同时解决了粒子PHD滤波器实现过程中出现的问题。本课题将深入开展基于随机集的改进粒子PHD滤波器的理论研究,具体研究内容包括:(1)建立适于粒子PHD滤波器的视频目标运动模型和量测模型及其噪声分布;(2)建立改进的粒子PHD滤波器,研究粒子PHD滤波器在视频跟踪中存在的缺陷,包括不能形成多目标的航迹,不能跟踪位置未知新生目标, 以及状态抽取不可靠的问题,理论分析并推导改进的粒子PHD滤波器;(3)最后,结合粒子PHD滤波器的实现条件,分析影响粒子PHD滤波器运行效率的因素,设计粒子PHD滤波器的实现方法,保障粒子PHD滤波器在视频跟踪系统中快速稳健实现。研究成果可以直接用于视频目标跟踪系统及相关领域的应用要求,具有重要工程价值。
中文关键词: 随机有限集;视频跟踪;概率假设密度;粒子滤波;状态估计
英文摘要: A new algorithm of multi-target visual tracking based on random finite set (RFS) and its implementation method are proposed, namely the improved probability hypothesis density (PHD) filter under RFS framework. The proposed algorithm is able to overcome the limitations of the original particle-PHD filter in application to visual tracking, and is capable of dealing with the difficulties of implementing particle-PHD filter in a real visual tracking scenario. The research contains: (1) design the dynamic and observation models with rational noise probability distributions for particle-PHD filter; (2) propose the improved particle-PHD filter to overcome the limitations of the original particle-PHD filter in tracking visual targets, which cover providing no information on the identity of targets and losing tracks of new born targets appearing at arbitrary positions;(3)given the required conditions in implementing particle-PHD filter and ingredients related to the efficiency of running the particle-PHD filter, design an effective implementation method for particle-PHD filter, achieving a fast and robust visual tracker based on particle-PHD. The research will provide new methods and theories for visual tacking systems and its related practical applications.
英文关键词: Random finite set;Visual tracking;Probablistic hypothesis density;Particle filtering;State estimation