【作者】 奚慧婷；

【导师】 黄昶；

【作者基本信息】 华东师范大学 ， 通信与信息系统， 2008， 硕士

【摘要】 运动目标跟踪在军事视觉制导、机器人视觉导航、工业产品检测、医疗诊断、交通监视等领域均有重要的实用价值和广阔的发展前景。由于被跟踪目标本身特征的多样性、所处环境的复杂性,使目标跟踪成为挑战性的课题,特别是跟踪过程中出现的遮挡问题成为限制跟踪算法鲁棒性的关键因素。本文主要针对刚性运动目标的跟踪问题进行研究,详细讨论了基于边缘特征的刚性运动目标跟踪算法。在该基础上,又进一步讨论遮挡情况下的边缘匹配和灰度匹配算法,论文的主要工作有:1、刚性目标的轮廓线或边界线是视觉感知与理解的重要线索或特征,采用帧间图像差分法进行运动目标检测,利用随机边缘特征有效地对目标进行跟踪,该算法对目标的视角、光线及运动状态的变化等不利因素有较好的适应性。2、针对刚性运动目标跟踪中的遮挡问题,提出边缘匹配和灰度相关匹配的跟踪算法。目标灰度单一时采用边缘匹配算法,目标灰度丰富时采用灰度相关匹配算法。对目标自适应分块,使各子块具有较明显的特征,增加遮挡判定和子块匹配的置信度;然后通过各子块来准确判定被遮挡区域,在遮挡情况下利用剩余的未被遮挡的子块进行灰度相关匹配和表决,在目标被完全遮挡情况下利用Kalman预测进行跟踪。该算法对于刚性目标的跟踪具有很好的实时性和跟踪精度。更多还原

【Abstract】 Moving target tracking plays an important role in military visual missile guidance, robot vision navigation, industrial product inspection, medical diagnosing and traffic surveillance, etc. Due to the facts of variety of the feature of the targets, the complexity of the environmental background, moving target tracking is a challenging subject. Especially the occlusion problem in the course of tracking has been one of the crucial restrictive factors to the robustness of the tracking algorithms.This thesis mainly focuses on the tracking of rigid moving targets. It detailedly discussed the tracking algorithm based on edge-feature-matching, particle filter and intensity-correlation-matching algorithms under occlusion.The major contribution can be summarized as follows:1. The contour line or the border line of the rigid target is the important clue and feature for visual sense. Two consecutive frames subtraction is applied to detecting the moving objects and random edge feature is contributed to object tracking. The algorithm can adapt well to the disadvantageous factors such as angles of target point, rays, and moving states of targets.2. Aiming at the general tracking of rigid targets under occlusion, we proposed a tracking algorithm based on edge and intensity matching. Edge matching algorithm is applied when the intensity of target is uniform, while intensity matching algorithm is applied when the intensity information is abundant. The target is adaptively divided in order to make each block with distinct feature to increase the accuracy degree of occlusion estimating and block matching. Then through blocks, the occlusion regions are accurately estimated, and match and vote are made using the remaining un-occluded blocks. Kalman prediction is used to track when full occlusion happens. This algorithm can track rigid target accurately and real-time.更多还原