项目名称: 目标运动突变和几何外观急剧变化的视觉跟踪
项目编号: No.61273273
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 陆耀
作者单位: 北京理工大学
项目金额: 81万元
中文摘要: 视觉跟踪是计算机视觉研究领域中的一个重要问题。运动和几何外观急剧变化情况下的目标跟踪是视觉跟踪的难点,基于连续运动条件假设的传统视觉跟踪算法在解决这一问题时有很大的局限性。 本项目提出了在目标运动和几何外观剧急剧变化情况下的视觉跟踪算法研究,包括:1)在贝叶斯滤波框架下,针对目标运动的不连续性进行统计建模, 设计有效的模型更新方法,从而对目标的空间区域按照不同的置信度进行有效的分类;2)针对目标的几何外观急剧变化进行自适应建模,既保留目标的空间结构信息又能体现其外观的几何变化,实现在复杂条件下的自适应目标跟踪;3)研究序贯马氏链蒙特卡洛(adaptive MCMC)跟踪算法,解决陷入局部最优状态的采样问题,实现对目标运动和几何外观急剧变化的滤波分布的自适应采样。最后,基于公用测试数据集对跟踪算法的性能和采样算法的稳定性进行测试,并进行算法的计算复杂度和收敛性质的理论分析。
中文关键词: 视觉跟踪;突变运动;姿态估计;视频对象分割;
英文摘要: Visual tracking has received worldwide attention in computer vision community. One of problems with challenge is the tracking of object with abrupt motion and geometric appearance change. The traditional visual tracking methods cannot be available because of their limitation of continuous motion assumption. In order to solve this tracking problem, we propose a new tracking method including: 1) statistically modeling uncontinuous motion in the framework of Bayesian filtering and designing the updating method of the model, which is able to make effectively classifying object area according to the different dependability. 2) adaptively modeling object with abrupt geometric appearance change, which is able to both keep the space structure of object and reflect geometric appearance change which is useful for adaptively tracking of nonrigid object. 3) researching Aaptive Markov Chain Monte Carlo (AMCMC) sampling based tracking method to overcome the local-trap problem in sampling. Finally, we examine the performance of the proposed tracking and sampling methods with the public dataset and make analysis of computational complexity and convergency of the proposed tracking method.
英文关键词: visual tracking;abrupt motion;pose estimation;video object segementation;