【作者】 李安平；

【导师】 敬忠良；

【作者基本信息】 上海交通大学 ， 控制理论与控制工程， 2007， 博士

【摘要】 视频目标跟踪是计算机视觉领域的一个核心问题,在军事制导、视频监控、机器人视觉导航、人机交互、以及医疗诊断等许多方面有着广泛的应用前景。视频目标跟踪的研究目的是模拟人类视觉运动感知功能,赋予机器辨识序列图像中运动目标的能力,为视频分析和理解提供重要的数据依据。视频目标的跟踪往往由于复杂的背景图像和目标本身的运动变得非常困难。尽管人们对视频目标跟踪进行了较广泛的研究,并提出了许多有效的跟踪方法,但是针对复杂环境下的视频目标,开发出一套鲁棒的跟踪算法仍存在较多困难。本文针对复杂环境下的视频单目标和多目标跟踪问题进行了重点研究。对单视频目标跟踪时,重点研究了目标观测模型的设计;对多视频目标跟踪时,重点研究了目标在场景中出现和消失、目标具有相似外表、目标之间交叉运动和相互遮挡等问题。本文的主要研究成果如下:1.针对复杂环境下的视频目标,提出了一种多特征自适应融合的视频单目标跟踪算法。在该算法中,目标的观测由多种特征的融合信息描述。在对每个特征信息进行融合时,采用了基于模糊逻辑的融合策略,模糊逻辑根据当前的跟踪环境自适应调节各特征信息的权重,从而实现各特征信息间的自适应融合,增加了描述目标观测的可靠性,提高了目标观测模型的鲁棒性;在跟踪目标时,采用了概率粒子滤波算法,将多特征信息自适应融合的观测模型结合到概率粒子滤波算法中,实现了较复杂环境下的视频目标跟踪。2.提出了一种基于自适应表面模型的视频单目标跟踪算法。该表面模型在跟踪期间能适应目标表面的缓慢或快速变化。在该模型中,每个像素的灰度值随时间的变化过程由一混合高斯分布描述。为了适应跟踪期间目标表面的变化,这些高斯参数通过EM算法在线更新;在对目标进行跟踪时,设计了基于自适应表面模型的目标观测模型,并将该观测模型结合到概率粒子滤波算法中去;针对目标发生部分和完全遮挡问题,我们通过采用一种鲁棒估计技术来降低被遮挡部分的像素对目标观测似然的影响以及对表面模型更新的影响。以上这些措施大大增加了复杂环境下视频目标跟踪的鲁棒性。更多还原

【Abstract】 Visual target tracking is a key problem in computer vision, it has a wide range of applications in military guidance, visual surveillance, visual navigation of robots, human-computer interaction, and medical diagnose, etc. The goal of visual target tracking is to imitate the motion sensibility of human vision, empower the machine with the ability of perceiving the target motion in the image sequence, and provide an important data source for visual analysis and understanding. Visual target tracking often becomes very difficult due to complex image backgrounds and the target motion. Although visual target tracking has been widely researched and many effective algorithms have been proposed, there are still a lot of difficulties in tracking the targets in complex environments.In this dissertation, the research is focused on the tracking problems of single target and multiple targets. For single-target tracking, the work is focused on the design of the observation model. For multi-target tracking, the work is focused on the problems of the appearance and disappearance of targets in scene, the similar appearance of targets, the cross movement of targets, and occlusion, etc.The main contributions of this dissertation are summarized as follows:1. For the visual target in complex environments, a tracking algorithm based on the multi-cue adaptive fusion has been proposed. In this method, the target observation is represented by multiple cues. When fusing each cue, a fusion scheme based on fuzzy logic has been developed. The fuzzy logic adaptively adjusts the weight of each cue according to the current tracking situations, so each cue is adaptively fused during tracking, which increases the reliability of observation and improves the robustness of observation model. When tracking targets, the probabilistic particle filter has been adopted, and the更多还原