【作者】 宋新；

【导师】 沈振康；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2007， 博士

【摘要】 地地战术导弹是打击防区外地面军事目标的重要武器之一。由于目前大部分地地导弹使用惯性导航,其圆概率误差很难达到米级的攻击精度,对于地面目标的攻击效能很低,特别是面积较小的目标,因此迫切需要末制导技术提高其命中精度。红外末制导技术是提高导弹命中精度的关键技术之一。本文以红外末制导成像导引头课题为背景,对复杂背景条件下的地面目标的检测、识别与跟踪技术展开了研究,并对导引头的系统实现方案进行了设计。论文的主要内容如下:1.研究了噪声和干扰条件下的非线性边缘检测技术,提出了具有边缘方向信息的柔性形态滤波边缘检测算法,并且提出了适合硬件实现的基于快速中值滤波与加权差分的边缘检测算法;2.针对红外图像中机场跑道的目标特征,提出了改进的斜率截距变换检测机场跑道的方法;通过统计边缘方向减少了参数空间转换,利用机场跑道的指向性特征进行峰值提取;提出了基于边缘连接性和知觉编组的线段提取方法,通过边缘的连接性建立线段列表,然后用知觉编组的方法进行进一步处理,实现了对桥梁目标的检测、识别。3.分析了红外图像中雷达天线罩的目标特征;提出了基于曲率估计的雷达天线罩目标检测方法;采用基于梯度约束的边缘跟踪,然后进行曲率近似计算并分段,然后对分段后的弧线进行计算得到圆参数;研究了基于核函数直方图的Mean Shift目标跟踪方法,提出了基于增强直方图的Mean Shift目标跟踪算法,实现了对雷达天线罩目标的稳定跟踪。4.研究了红外图像建筑物目标的检测、识别和跟踪方法:提出了改进的快速水平集方法,从数据结构实现、曲线进化的平滑处理和算法的结束条件等方面对算法进行了改进。实现了对建筑物目标的轮廓提取,并且通过形状轮廓识别的方法进行建筑物目标识别;另外,对于建筑物目标旋转的情况的跟踪,研究了基于带宽矩阵表示的高斯Mean Shift跟踪方法,实现了对建筑物的稳定跟踪。5.讨论了末制导导引头信息处理机的系统设计策略,根据末制导导引头的系统要求,设计了导引头信息处理机的实现方案;并且对硬件的详细设计进行了分析;并且对FPGA实现的主要模块和算法进行了仿真验证;对软件算法进行了层次划分,对不同目标的处理流程进行了详细的介绍。更多还原

【Abstract】 The ground to ground tactical missile is one of the important weapons that attack ground target located outside of defense area. A majority of ground to ground tactical missiles are depending on inertial navigation system, so CEP can hardly reach meter-level precision and result in low battle efficiency, especially for small area of target. Terminal guidance technology is imminently needed to improve lethality of the missiles. Infrared terminal guidance is a key technology to improve lethality of the missiles. Taking infrared image terminal guidance as main task, this dissertation investigate detection and tracking of ground target in complex background, and design the information processor of infrared terminal seeker.The main research contents of the thesis list as follows:Firstly, technologies of no-linear edge detection in noise and disturb condition is researched. An soft morphological edge detection method with edge direction is proposed. Then, a method of edge detection based on fast median filter and weighted difference is presented. This method is feasible for hardware implementation.Secondly, according to characteristics of runway in infrared image , an runway detection method based on improved Slope-Intercept Hough Transform is proposed, in which statistics of edge direction is used to reduce parameter space conversion and direction of runway is used to extract maximum of parameter space. A line segment detection method based on connectivity of edge and perceptual organization principle is proposed. Line segment lists is builded and dealt with perceptual organization. Bridge detection and recognition is achieved by this method.Thirdly, characteristics of radar radome in infrared image is analyzed. A new circle detection method based on curvature estimation is proposed. In this method, after boundary tracking based on edge direction, boundary is segmented by estimated curvature, then circle parameter is computed in segmented boundary. Mean Shift tracking based on kernel histogram is studied, then an improved Mean Shift algorithm based on new histogram is proposed. This method can stably track radar radome in complex background.Fourthly, methods of building detection, recognition and tracking are researched. An improved FLS(fast level set) method is presented which is improved by new data structure, end condition and smoothness of curve evolution. Building is extracted by this method and recognized by shape matching. For tracking rotating building target, a method based on Gaussian Mean Shift with bandwidth matrix is studied.Fifthly, system processing strategy is discussed. Architecture of infrared terminal seeker is designed and detailed circuit design is analyzed. System flow of software is analyzed in detail. Implementation of fast median filter and weighted difference algorithm is simulated. Fifthly, system processing strategy is discussed. Architecture of infrared terminal seeker is designed and detailed circuit design is analyzed. FPGA Implementation of main modules and edge detection algorithm is simulated. Algorithm is hiberarchy partitioned.Processing flow of every target is analyzed in detail.更多还原