【作者】 张杰；

【导师】 孙燕；

【作者基本信息】 南京师范大学 ， 计算机应用技术， 2016， 硕士

【摘要】 运动目标跟踪技术是计算机视觉的重要部分,在运动目标跟踪、运动状态理解、运动轨迹发现中具有重要的研究价值和意义。在安全监控、商业智能、交通监管、医学应用、生物搜索以及虚拟现实等很多应用领域都有着广泛的应用前景。本文研究基于Mean Shift算法的运动目标跟踪(简称为MS)。针对MS算法中存在的不足,在目标跟踪中的模板特征表示、模板更新问题以及尺度自适应等方面进行了研究,提出了改进和创新方法,获得了较好的跟踪效果,论文取得了主要创新成果如下：1.提出了精简等价模式与颜色融合的MS目标跟踪的方法。该方法针对模板特征表示单一以及颜色描述时结构信息缺失的问题,设计了综合等价模式和微结构中的次要模式优势的精简等价模式,保留了LBP等价模式的降维特性,删除了对识别精度影响小的微结构中的次要模式,实验结果表明：本文提出的方法能延长跟踪,降低了平均中心位置错误误差,提高了目标跟踪精度。2.提出了目标模板在线更新的MS目标跟踪的方法。针对在复杂环境中目标模板漂移引起的以及遮挡更新引起目标跟踪丢失问题,本文设计了线性加权的模板更新以及避遮挡的目标模板在线更新算法,有效地分配权重以及分块预测遮挡问题,实验结果表明：本文提出的方法能够及时调整或更新模板,达到提高模板的准确度目的。3.提出了目标尺度自适应的MS目标跟踪的研究方法。针对跟踪目标的尺度变化时MS算法自身缺乏尺度自适应策略的问题,本文设计了以权重因子为尺度的尺度修正自适应算法,该方法利用了图像权重因子与目标窗口以及Bhattacharyya系数之间的相关性,将图像权重因子转换为求目标面积问题,并兼顾目标尺度变化的趋势而提出的修正方法。实验结果表明：本文提出的方法较好地适应目标尺度产生的变化。更多还原

【Abstract】 The tracking technology of moving target is an important part of computer vision, it has important research value and significance in the moving object tracking, motion understanding and trajectory of discovery. It has broad application prospects in the fields of security monitoring, business intelligence, traffic monitoring, medical applications, biological search, virtual reality and many other applications.In this paper, it is researched about the tracking algorithm of moving target based on the algorithm of Mean Shift (abbreviated as MS). In order to solve the existing problems of the MS, this paper studies on the aspects of the feature of target tracking template representation, the problem of updating template and the method of adaptive scale, and proposes improvements and innovative approaches, obtains a better tracking performance. The major innovations of this paper are as follows:1. This paper presents a target tracking algorithm based on MS merged by simplified Uniform Pattern and color feature. Considering that the feature of target tracking template representation is too single and structure information is missing when using color description, this paper designs a simplified Uniform Pattern combining advantages of Uniform Pattern and minor pattern of micro structure, reserves dimensionality reduction characteristic of LBP Uniform Pattern, and deletes minor patterns which have a little effect on recognition accuracy when using micro structure pattern, experimental results show that the algorithm in this paper can extend tracking time, reduce errors of the drifting center position and improve the target tracking accuracy.2. This paper presents a target tracking algorithm under the MS framework based on the online updating target template. Considering that the problem of the losing object tracking, causing by the drifting target template and occlusion in the cluttered dynamical background, this paper designs an online updating algorithm based on linear weighted updating template and anti-occlusion target template, effectively assigns weights and predicts occlusion by fragments. Experimental results show that the algorithm in this paper can adjust or update template timely and achieve the goal of improving the accuracy of object templates.3. This paper presents a MS target tracking algorithm with scale adaptation. Considering that MS algorithm is short of scale adaptation strategy, this paper designs a scale revised adaptation algorithm which regard weighting factor as scale. the method uses image weighting factor, target window and the correlation of the Bhattacharyya coefficient, converts image weighting factor for the problem of resolving target area, and it also consider the trend of target scale change. Experimental results show that the algorithm in this paper can adapted the target scale change better.更多还原