基于Mean Shift算法和Particle Filter算法的目标跟踪

【作者】 李涛；

【导师】 任明武；

【作者基本信息】 南京理工大学 ， 计算机应用技术， 2007， 硕士

【摘要】 目标跟踪技术在自动控制、监控技术、医学图像识别等领域的应用中有着独特的优势，但是近年来，目标跟踪技术仍然不能达到人们满意的效果，严重阻碍了其应用推广，随着硬件技术的飞快发展，动态图像的分析和理解成为研究的热点，目标跟踪技术应用在许多重要领域，使得该技术成为一个重要的研究课题。论文讨论了目前两种经典的目标跟踪算法：Mean Shift算法(均值偏移)和粒子滤波算法(Particle Filter)，分析了两种算法的特点；针对目标跟踪鲁棒性不高的问题，分析了用运动目标检测提取目标运动特征的技术，通过增加对目标特征描述信息，提高跟踪健壮性，并在以颜色直方图描述颜色特征的基础上，融合了目标的运动特征，设计了一种基于运动特征和颜色特征多特征融合的粒子滤波跟踪方法。论文对颜色直方图进行了分析，比较了颜色直方图和二阶直方图的优劣，结合了运动特征提取方法和扩大搜索范围的方法，给出了以均值偏移为理论基础，用二阶直方图描述目标颜色特征，实现目标跟踪的技术；针对粒子滤波存在的缺点，用二阶直方图描述颜色特征，设计了均值偏移和粒子滤波相结合的目标跟踪技术，使每个粒子表示状态更合理，在遮挡时能够实现很好的跟踪。更多还原

【Abstract】 Object tracking have special advantage in the automatic control, scout system,medical science picture identify etc. In recent years, Target tracking technique still can’tattain people’s satisfaction, seriously obstructed its application expansion, so it is an urgentproblem to research target tracking. The difficult problem in visual tracking is performingfast and reliable matching of target from frame to frameThe thesis explores currently two kinds of targets tracking algorithm, Mean Shiftalgorithm and Particle Filter algorithm. They are very good algorithm in visual targettracking area. The thesis explores the Detection of Moving Targets, To improve therobustness of visual tracking in complex environments, a novel tracking method based onadaptive fusion and particle filter is proposed, the image color and moving cues areadaptively fused to represent the target observation.The thesis explores an image description method based on second order histogramand increase searching scope to improves robustness of visual tracking. An algorithmbased on kernel histogram particle filter is studied, the dimension of particle is reduced andthe required particle is very few, New algorithm makes the best of middle value of particlefilter, so that the complexity of algorithm is not added.更多还原