基于在线学习的自适应粒子滤波视频目标跟踪  

Video Object Tracking Based on an Online Learning Adaptive Particle Filtering

作　　者：席涛[1] 张胜修[1] 颜诗源[1] 徐晓淼 

机构地区：[1]第二炮兵工程学院303教研室,西安710025 [2]第二炮兵驻211厂军事代表室,北京100076

出　　处：《光电工程》2010年第6期29-34,共6页Opto-Electronic Engineering

基　　金：学院青年科技创新项目(XY2009JJB27)

摘　　要：传统的粒子滤波视觉跟踪算法采用固定模型和大量粒子表征目标后验概率,不能满足复杂条件下的视频目标实时跟踪。为了提高跟踪的鲁棒性和稳定性及计算效率,本文提出将自适应状态演化方程和在线增量学习观测似然模型嵌入到粒子滤波算法;并采用在线自动调整粒子数目的策略,提高粒子滤波视觉跟踪的计算效率。室内外实验结果表明,文中提出的视觉跟踪算法不仅能准确、高效地跟踪序列图像中的运动目标,而且对光照、姿态变化引起的目标表观变化具有良好的鲁棒性。Particle filter has already been extensively applied to video object tracking, however the traditional visual tracking approaches based on particle filter, which employ an experiential state transition equation and observation likelihood model predefined in advance, cannot satisfy the request of the real time complex situation video object tracking In order to improve the robustness and stability as well as the computation efficiency of the video tracker based on particle filter, the adaptive state evolution equation and an online increment learning observation likelihood model are embedded into the particle filter, and the strategy for online self-adjusting the number of particle is adopted to enhance the computation＇efficiency. The experimental results show that the approach proposed in this paper not only track the moving object in the video accurately and effectively, but has nice robustness to the appearance variation caused by illumination and pose changes.

关 键 词：视觉跟踪 自适应模型 增量学习 子空间 粒子滤波 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]