摘要
提出了一种视觉跟踪任务中基于局部特征和概率图模型的目标建模方法,将目标表示为一组具有仿射不变性的区域特征,并通过概率图模型描述特征之间的空间约束关系。在目标跟踪过程中,首先在空域上利用信任传播算法,推断概率图模型中各个特征的状态,然后根据推断的结果设计改进的重要性采样函数,采用粒子滤波算法在时间域上对目标进行跟踪。为了适应目标在运动中的变化,模型根据特征的稳定程度自适应地进行更新。实验结果表明,该方法具有较强的鲁棒性,能够有效实现复杂场景下的目标跟踪。
A novel method of object modeling for visual tracking based on local features and probabilistic graphical model is proposed.The tracked object is represented using a collection of regional affine invariant features,among which the spatial constraints are described by a probabilistic graphical model.During tracking,belief propagation algorithm is first applied to infer the state of each feature in spatial domain.Then,the inferred results are employed to construct the proposal sampling function,by which a particle filter is adopted to estimate the target state.To adapt to changes in object appearance,object model will be updated adaptively according to the stability score of the features.The experimental results show that the proposed method can get reliable track even under complex real world conditions.
出处
《光电子.激光》
EI
CAS
CSCD
北大核心
2010年第1期124-129,共6页
Journal of Optoelectronics·Laser
基金
国家自然科学基金资助项目(60705005)
国家自然科学基金民航联合基金重点资助项目(60736046)
教育部博士点基金资助项目(20070610031)
关键词
视频目标跟踪
仿射不变特征
概率图模型
信任传播
粒子滤波
visual object tracking
affine invariant feature
probabilistic graphical model
belief propagation
particle filter