【作者】 赵丹培；

【导师】 王延杰；

【作者基本信息】 中国科学院研究生院（长春光学精密机械与物理研究所） ， 光学工程， 2006， 博士

【摘要】 本论文从两个方向研究了频域下的目标跟踪测量方法。一种是基于傅里叶变换相移特征的运动分析方法，它能够准确地估计出运动目标的平移和旋转参数矢量。文中在传统的相位相关算法基础上提出了几种改进的运动参数估计方法，其中基于SVD分解的相位相关算法不仅在计算量方面优于其它算法，而且在抑制噪声和抗遮挡等方面更显示出良好的性能，同时为满足系统实时性的要求，采用一种将二维图像映射为一维信号的相位相关简化算法，速度是传统算法的4倍；为实现运动矢量的亚像素匹配，提出一种基于频域相位差信息的相位差算法，它建立起一种空域下目标位置变化与频域下相位角变化之间的简单线性对应关系，速度比传统相位相关算法提高了8倍。用外场图像做了相应的仿真实验，取得了很好的实验结果，证明了算法的有效性和实用性。 另一种频域下的目标跟踪方法是利用小波变换局部模极大值的边缘检测方法，并结合小波多分辨分析的特性在去噪和图像增强等方面的优势，实现3％对比度下的微弱目标检测。文中比较了三种小波局部模极大值边缘检测方法在抗噪性、计算量及检测效果等方面的性能以及如何针对图像特点选取小波基，并提出一种针对不同背景的自适应阈值法，在抗噪的同时能够检测出更多的细节边缘。通过仿真实验证明了利用B样条小波的模极大值边缘检测算法对空中目标、模糊目标和低对比度目标都有很好的检测效果。也探讨了把利用小波分析的图像去噪和图像增强方法融入到边缘检测中，在抑制噪声的同时，从模糊图像中提取出目标边缘，将目标准确定位。更多还原

【Abstract】 In this paper, the methods of target tracking in frequency domain are studied from two aspects. One is the motion analysis based on phase difference of Fourier transform. It can accurately estimate the motion vector of object by identifying the displacement and rotation between two similar images. The paper proposed some improved methods to estimate motion parameters based on traditional phase correlation method. The method based on SVD is robust when noise present or object is sheltered because of using phase information in the frequency domain whose speed is nearly three times as quickly as original method. To satisfy the requirement of real time image processing system, a reduced method is presented which projects two-dimension image to one-dimension signal; thus, its velocity is increased fourfold. In this paper, a new method is proposed to estimate spacial object motion which based on phase difference in the frequency domain. It establishes a linear relation between positional movement of the object in spacial domain and phase change in frequency domain, whose speed is eight times as quickly as phase correlation arithmetic. It is fit for real-time image processing system because of sub-pixel precision and speediness. The simulated experiment data show that the proposed methods give a better detect result for the spacial object, which demonstrates the utility and practicability of these methods.The other method of target tracking is to detect image edges according local modulus maxima of wavelet transform based on multi-scale properties of wavelet更多还原