视频场景杂波背景移除技术研究

【作者】 张翔；

【导师】 李在铭；

【作者基本信息】 电子科技大学 ， 通信与信息系统， 2006， 硕士

【摘要】 以光学信号处理技术为基础的目标检测和跟踪技术是目前许多先进武器系统的关键技术,被广泛运用到多种装备中。红外及可见光成像检测和跟踪系统由于其处理的是多维信号(二维时间,一维空间),且是一种被动检测技术,因此具有隐蔽性好,抗干扰能力强等诸多优点.近年来,军事领域的要求不断提高,要求尽早最大限度的发现目标,使得序列图像中微弱目标检测和跟踪技术成为了研究热点。微弱目标的检测和跟踪问题可以分成三个部分:1.图像预处理;2.目标检测;3.目标跟踪。由于图像传感器的抖动和背景杂波的影响,在目标检测和跟踪之前有必须实行图像预处理,这正是本文的研究点。本文的主要工作在于给出了微弱目标的数学定义,从人眼视觉和机器视觉两方面严格界定了微弱目标,建立了基于背景预测的理想背景杂波抑制系统并分析了其性能,从信噪比,高斯性,白化性三个方面研究了目前文献上出现的大部分残余背景评价指标并给出了详细的改进实现方法。在这些研究的基础上,开展了具体的杂波抑制技术的研究。首先分析了经典的非参数法,对于四种具有代表性的核,从前述的三个性能评价方面做了分析和对比,指出了其速度快的优点和对非平稳图像适应性差的弱点,针对非参数法的弱点,重点研究了对非平稳图像适应良好的卡尔曼杂波抑制技术:建立了非平稳图像的类自回归模型,在此基础上建立了二维卡尔曼滤波基础的两个方程:状态方程和测量方程;建立了非平稳图像准平稳区域快速划分算法:基于四叉树法的有限分裂合并算法;二维空间的基于K排序的滤波路线算法,突破了空域滤波路线上区域相邻的限制;在这些研究的基础上实现了快速卡尔曼估计,实验验证了该方法相对逐点卡尔曼估计可以提高运算速度三倍左右;杂波抑制结果表明传统的高斯性检验并不适合卡尔曼估计后的残余图像,由此建立了残余图像的双参数拉普拉斯模型,实验表明其可以完好的吻合残余图像的概率密度曲线。更多还原

【Abstract】 The target detection and tracking technology based on optical signal processing is one of the most important technologies applied in many advanced weapons. Since the optical detection and tracking technology deals with 3D space-time signal and is a kind of passive detection technology, it has a lot of features, such as concealment, anti-jamming and high precision tracking performance. At present, the pressure from military and others demands that target should be detected as soon as possible and as far as possible, so there are a great of research interests in studying of the detection and tracking technology of dim point moving target from image sequence.The dim target detection and tracking system can be divided into three parts, the image pretreatment, the target detection and the target tracking. Because of the influence of the wobble of the sensor and the clutter background, the pretreatment is necessary and also it’s the research trend of this dissertation.The main contribution of the dissertation is the whole studying of the pretreatment technology of image sequence including dim point moving target. In this dissertation, the research trends for the problem have been introduced; the‘dim’and‘point’has been strictly defined in mathematics from machine vision and human vision; the ideal clutter suppression system based on clutter predication and the realization and evaluation of evaluation index has been studied, in succession the clutter suppression technologies have been researched. Firstly, the classic nonparametric algorithm has been analyzed in detail and systematically, for it’s weakness that it cannot remove the non-stationary clutter ideally, Kalman filter algorithm for clutter suppression in 2D image signal has been built. Secondly, Fast Adaptive Kalman Filter is presented based on fast wide-sense stationary areas partition algorithm: limited combination and division algorithm based on quarti-tree algorithm, new taxis filter route algorithm which can break through the limitation of the necessity of pixel neighborhood of 2D filter and Laplace data model with two parameters which is perfectly suitable for the residual image of Kalman clutter suppression. Experiments show that the performances of fast adaptive Kalman filter is good.更多还原