【作者】 龚俊亮；

【导师】 何昕；

【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， 2013， 博士

【摘要】 随着常规兵器朝着一体化、信息化和远程精确打击方向发展，装备信息化、平台自动化、侦查远程化、弹药制导化、系统网络化、毁伤多样化等趋势和特点日益明显，上述变化均对常规兵器试验鉴定提出了新的更高要求。作为试验鉴定的重要组成部分，通过测量目标弹体飞行弹道参数来揭示被测试武器装备内在性能和固有特性已成为靶场采用的最直接最有效的手段，且其面临着愈加严峻地挑战。在靶场光学测量设备中，常用的可见光成像系统拍摄的可见光图像包含清晰的弹体信息，通过判读可以得到弹体飞行轨迹和姿态参数，但可见光成像容易受天气环境影响，当目标弹体与成像背景颜色相近时，很难进行目标弹体的识别，这些局限性导致在试验中不能保证拍摄到的可见光图像可以用于完成对目标弹体飞行弹道参数的精确测量。近些年来，随着红外技术的不断发展，红外测量以其捕获能力强、作用距离远、对天气和背景的适应性较好等特点，越来越受靶场光测的青睐。可当目标为远距离成像时，其在靶面的成像占据很小的像元数量，且目标信噪比低，容易被背景淹没。因此，红外弱小目标的检测跟踪问题一直是机器视觉领域的难题。同时导弹及飞行器甚至炮弹的运动速度及机动性能越来越强，尤其是当视场中为多目标时，检测与跟踪能否实时实现也关系到系统性能的高低。本文的研究内容可为我国靶场红外测量系统的研制提供相应的理论研究基础和技术支持。结合在研国防科研项目需求，本文探讨了靶场常见的天空背景下的红外弱小多目标的检测与跟踪问题。遵循传统的研究步骤，本文将该问题分为图像预处理、弱小多目标检测与多目标跟踪三个部分进行研究。在深入分析国内外红外弱小多目标检测跟踪研究现状和研究进展的基础上，在图像预处理部分，本文将图像多尺度几何分析的最新研究进展引入红外图像预处理，提出了采用剪切波变换的红外图像预处理方法。多目标检测方面，结合弱小目标描述模型，提出了基于尺度空间的红外弱小多目标检测算法，用于在复杂背景下快速而准确地检测出多个弱小目标。在多目标跟踪方面，提出一种结合改进的辅助粒子滤波与马尔科夫随机场的多目标跟踪算法，将多目标的跟踪问题转换为图模型的推理问题。同时设计了基于多DSP处理器的多目标跟踪器硬件平台，对移植后的算法进行了多项实验测试，验证了算法与硬件平台系统方案的正确性和可行性。更多还原

【Abstract】 With the development of conventional weapons heading towards integration,informationization and remote precision strike, the trends and characteristics aboutinformationization of equipments, automation of platform, remote implement ofdetection, navigation of ammunition, networking of systems, diversify of devastationhas become increasingly evident, these changes mentioned above are proposing newand higher demands on conventional weapons tests. As an important part of the test,measurement of the target missile flight trajectory parameters is the most direct andeffective means used by a firing range to reveal the intrinsic properties and theinherent characteristics of the weaponry, and now this method is facing increasinglyserious challenges. Visible light imaging system is a kind of commonly used opticalmeasurement equipment in a firing range. The visible light images often containclear information about the targets, the flight path of the missile body and its attitudeparameters can be precisely obtained through interpretation, but the images arevulnerable to the impact of weather conditions, and when the targets body imagingsimilar to the color of the background, it is difficult to identify the target from thebackgrounds. With these limitations, it can not guarantee that the visible light imagescan be used to complete the accurate measurement of the flight trajectory parametersof the target missile body.In recent years, with the continuous development of infrared technology, it is increasingly favored by the firing range to use infrared image because of its distancemeasurement, the noticeable ability of capture, and fine adaptability to the weatherand backgrounds. However, when the targets have a far distance with the infraredoptical measure equipment, the image of the targets are small, with weak signals,and are submerged in complex backgrounds, all of these make the detection andtracking of the targets very difficult. Therefore, the question of how to detect andtracking the dim and small targets in complex backgrounds is becoming a technicaldifficulty today. On the other hand, considering the system performance index points,especially when the number of target is larger than one, the real-time implementationis more take pains. so, it is visible that the research on real-time infrared smalltargets detection and tracking not only has an important theoretical significance, butalso has significant practical value.Based on national defense scientific research needs, this paper discusses thedetection and tracking problems of infrared dim small multi-targets under complexbackgrounds. Following the traditional research steps, this paper divide the probleminto three parts study, the pre-processing of image, the detection of the dim smallmulti-targets and targets tracking. Based on the deep analysis of this problem athome and abroad, in the part of pre-processing of image, the latest researchachievement of image multi-scale geometric analysis---Shearlet Transform isintroduced, and based on that theory, a new method in image pre-processing isproposed; To solve the second part of the problem---the detection of small dimtargets, a method based on scale-space theory was proposed. In the last part, analgorithm combine improved auxiliary particle filter with Markov random field isproposed to solve the problem of targets tracking, and simulate multi-tracking modelby using Markov random field. Finally, a multi targets tracking hardware platformusing three DSPs is introduced, through a number of experiments this paper test thetransplantation of the algorithm, verify the correctness and feasibility of the systemprogram and algorithms.更多还原