【作者】 贺柏根；

【导师】 朱明；

【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） ， 机械电子工程， 2012， 博士

【摘要】 自从第二次世界大战首次出现以来，精确制导武器如雨后春笋般不断涌现，纵观近半个多世纪来世界上历次大大小小的局部战争，制导武器的使用比例在急速上升，从二战时的尝试性使用到越南战争中崭露头角，再到最近的利比亚战争中完全占据主导地位，大有取代常规武器的趋势。尽管目前来看两种武器各有特点和优势，但常规武器制导化已然成为世界军事发展必然趋势却是不争的事实。基于以上背景本文在对当前世界精确制导武器做大量分析研究和归纳总结的基础上，着重展开对以美国“标枪”为代表的短程电视末制导导弹自动目标识别技术的研究。“标枪”是一款肩扛式短程作战武器，发射者瞄准目标并发射后导弹可以自主完成目标识别、跟踪和攻击任务，做到了真正的“发射后不管”。对这一类型末制导导弹电视目标识别系统的研究国内尚不多，其难点主要有三个方面：首先，射手瞄准目标存储目标模板与目标向下攻击进行模板匹配时，导引头的成像会有一个比较大的视角变化，这一大视角变化是普通匹配算法难以解决的；第二，导弹向下攻击时有个点火过程，在这一短暂的过程中会发生剧烈抖动，该阶段背景对目标干扰比较严重这对图像跟踪算法的精确性和速度提出了很高的要求；第三，当导弹接近目标时目标尺寸会不断变大，这就要求该阶段图像跟踪算法对尺度变化有很强的鲁棒性。另外，在硬件实现方面，受限于肩扛式末制导到的体积和重量要求，需要在有限的硬件板卡尺寸下采用高速数字信号处理芯片保证运算的实时性。针对这些技术要求和难点本文展开研究并在Matlab7.5和VC++6.0工具对提出的算法进行了实验验证。本论文的主要工作可以总结为以下六点：1.在查阅大量国内外文献资料基础上对当今世界众多的精确制导武器进行了全面的分类归纳总结，对各类制导武器的技术特点及应用环境做了全面分析和介绍。并对当前的成像精确制导技术做了相关介绍。2.针对电视末制导导弹模板匹配过程中存在的大视角变化问题，在对多种基于特征点的抗仿射变换匹配算法比较后提出将改进的全仿射不变SIFT匹配算法应用至此，实验证明所提算法能够满足电视末制导导弹大尺度视角变化的要求。3.提出了一种Harris角点统计特性和PSO(Particle Swarm Optimization,PSO,粒子群优化)算法相结合的目标跟踪技术，集精度与速度于一体，以满足电视末制导导弹初期跟踪时目标在视场内变化剧烈的问题。最后用一组实战视频对所提算法的实时和有效性进行验证。4.针对电视末制导导弹后期跟踪时目标尺度变化较大的问题，提出了基于Mean-shift（均值漂移）的尺度自适应目标跟踪改进算法，通过对经典Mean-shift算法核阴影函数的改进，有效地解决了该算法对尺度变化目标跟踪的不足。实验证明改进后的算法对目标尺度变化有很好的实时性和鲁棒性。6.介绍了本文设计的基于DSP+FPGA图像处理器。分别对该处理器的原理、器件选型功能做了具体介绍。该硬件板卡在面积较小大芯片较多的情况下有效解决了电磁兼容等复杂问题，满足了工程实践的应用要求。最后给出了处理器的板卡及其安装平台实物图，并给出了该图像处理器在实际工程应用中的部分视频截图。本文所提的Harris角点统计特性和PSO算法相结合的目标跟踪算法及基于Mean-shift的尺度自适应目标跟踪改进算法都是在PC机上用VC++6.0进行仿真实验的，这两种算法都能满足实时性要求，可以将其往硬件板移植。所设计的图像处理器已经应用在了实际的工程项目中，只有改进的全仿射不变SIFT匹配算法计算复杂、计算量大，要实时工作仍需做进一步优化处理。更多还原

【Abstract】 Since it appeared for the first time in World War II, the precision-guided weapons have sprung up like mushroom. Looking back over every large and small local war in the world during the past more than half a century, guided weapons were used in the ground rising rapidly. From the first appearance in World War II to be used in the Vietnam War, and then occupied a dominant position in the recent Libyan war, there seems to be a substitute for conventional trends. Although the two arms have their own characteristics and advantages, but using guidance technology in the conventional weapons will be an inevitable trend for the world’s military development. Based on this background, in this article we will make a detailed analysis of the precision-guided weapons on the current world, and then do more detailed research in short-range TV-guided missiles whose representative are "Javelin" missile made by US."Javelin" is a shoulder-fired weapon for close range combats. After the missile-man aims at the goal and fired missiles, they can complete the task of target identification, tracking and attacking independently, it has achieved the true automatic target recognition. There is no relevant information on this type of TV-guided missile domestic; its difficulty mainly has three aspects:first, there will be a large view changes between images made templates stored before the missile fired and images used for template matching. It is difficult for traditional image matching algorithm to resolve this problem. Second, the missile has an ignition process before attacking down, a violent of shaking will occur in this short period, and background will have a serious interference on the target. So the image tracking algorithm must have high accuracy and speed. Third, the target will become larger and larger when the missile gets close to it, this requires the image tracking algorithm in this phase to have very strong robustness to scale changes. Moreover, in hardware designing, the size of circuit board is limited by the missile’s size and weight. So we should arrange the location of the high-speed digital signal processing chip appropriately and make sure it’s effectiveness. This article studied mainly based on these technical requirements and difficulty, and made the following arrangements:1. In the second chapter we will conduct a comprehensive classification and summary to precision-guided weapons in the world based on a large number of literature data domestic and abroad. Analysis of the characteristics and application environment to the various types of guided weapons will be presented in this chapter too. The last of this chapter has made an introduction to the current imaging precision guidance technology.2. The third chapter of this article is mainly to solve the problem of having large view changes in template matching for TV-guided missiles. After comparing many images matching algorithms of anti-affine transforming based on features, we propose to use an improved anti-entire affine SIFT matching algorithm to solve the view changing problem.3. We will presents a target tracking algorithm combined the statistical characteristics of Harris corners and PSO algorithm. This algorithm sets the accuracy and speed in one to meet the images’violent shaking during TV-guided missiles early tracking. At the end of this chapter, we use a set of actual combat video doing experiment to verification the timeliness and effectiveness of the proposed algorithm.5. Chapter5focused on the target’s size changes during the TV-guided missile’s last tracking, proposed scale adaptive tracking algorithm based on Mean-shift. The new algorithm solved the problem of mean-shift’s lack to the target’s criterion size changes. Experiments show that the improved algorithm is robust to scale changes and a real-time one.6. In the6th chapter of this paper, we mainly introduced the image processor based on DSP+FPGA. This chapter made a detailed introduction to the processor’s principle, chips’selection and their function respectively. The biggest problem in desing this circuit board is we should not only arrange many larger electric chip in a small area but also solve the complex electromagnetic compatibility. So that it can meet the application requirements in engineering practice. At the end of this chapter we gave out the physical images of the image processor’s circle board and its mounting platform flowed by many practical images in practical applications.The algorithms of target tracking based on statistical properties of harris corner and pso and mean-shift that is affine to scale changes are both simulated on a PC with VC++6.0. Both of them are meet the real time requirements and have been used in practice. Regrettably, the image matching algorithm based on SIFT should make more effort to real-time work due to its computational complexity and large amount of calculation.更多还原