【作者】 胡永生；

【导师】 陈钱；

【作者基本信息】 南京理工大学 ， 光学工程， 2008， 博士

【摘要】 红外预警系统(Infrared Surveillance System,IRSS)靠被动地接受目标的红外辐射来探测、识别和跟踪目标。它可以单独使用,也可以与其它预警设备如雷达系统交联使用,广泛用于军事目标的侦察告警或连接武器平台的火控系统。相比于雷达预警,红外预警系统由于具有隐蔽性好、角分辨率高、抗电磁干扰能力强以及体积小、重量轻、机动性强等优点,正逐渐成为现代防御体系的一个重要组成部分,受到各国军工部门的重视。在红外预警系统中,复杂背景中红外弱小目标探测与跟踪是一个关键技术。研究复杂背景中弱小目标的实时探测算法,可以扩展红外预警系统的作用距离,对于增加火控系统的反应时间、提高己方的生存概率具有重要的意义。但当目标距离较远时,目标在红外图像中表现为点目标,并且图像信噪比低,目标基本上被背景和噪声淹没,没有距离信息,此时无法利用形状、大小、纹理等特征来识别目标。单帧探测可能产生很多虚假目标,无法获得所要求的探测概率和虚警概率;在序列图像中,根据目标运动连贯性,利用时间上的信息来探测弱小目标,成为有效的解决途径。为提高现有系统探测能力,本文对复杂背景中红外弱小目标的探测与跟踪技术进行了深入研究。首先详细讨论了小目标在探测器上的成像过程,并建立了相关的数学模型。从得到的红外图像序列出发,对弱小目标在探测器上的成像特性,包括大小、形状、灰度、对比度和目标与背景的局部相关性等进行了分析。在此基础上,本文提出了一系列的算法来探测弱小运动目标:基于有限自适应邻域的红外图像预处理算法进行弱小目标增强和噪声抑制;设计了基于数学形态学的弱小目标探测算法和恒虚警率探测技术;为进一步降低单帧目标探测的虚警率,本文还提出了基于目标区域灰度概率分布函数匹配的小目标探测技术。在图像序列中利用目标运动连贯性的多帧跟踪探测技术,是提高复杂背景中弱目标探测概率的一个主要技术途径。本文对复杂背景中的弱小运动目标探测进行了深入研究,提出采用基于动态规划技术的多帧探测算法,主要依据帧间目标的相关性和目标运动特征的连贯性,构造了一种新的阶段值函数和惩罚函数。针对红外搜索预警系统,本文还提出一种基于多假设的目标跟踪方案。根据目标在探测系统中的运动特性,采用卡尔曼滤波实现目标的预测与滤波,提出了完整的航迹起始、维持和终结方案。提出了一种基于多目标特征综合评价因子的数据关联新技术,用目标坐标位置、灰度、大小和局部信杂比在时间上的连续性进行数据关联。设计了目标探测与跟踪仿真系统,仿真结果表明,本文提出的图像序列中弱小目标探测跟踪方案具有强的抗杂波、多目标跟踪、抗数据丢失和实时处理能力。根据对复杂背景中红外弱小目标探测识别与跟踪的理论研究和仿真结果,为低空红外预警系统研制了具有弱小目标实时探测能力的红外信号处理模块。实验验证了本文相关算法的有效性和实用性。更多还原

【Abstract】 Infrared Surveillance Systems (IRSS) are electro-optical detectors which detect, recognise and track targets by receiving their Infrared emission passively. The IRSS which, either working alone or as part of radar surveillance suite, search and scout military targets, can provide target information to fire control system of weapon platforms. In general, IRSS has passive detection and invisibility, high resolving ability, fine anti-jamming ability, small size, lightweight and high maneuverability. For these advantages, IRSS bcomes an important component of the nation defense system, and obtains wide attention and energetic cooperation in recent years.Infrared dim and small target detection and tracking in clutter environment is a key technique in IRSS. Efficient target detection and tracking algorithms for low signal-to-nose ratio (SNR) environment can result in longer alarm range of IRSS. It is of great significant to increase response time of fire control systems and survival probability. But when target is far away from sensors, the target is point-like and the image SNR is very low. The target signal amplitude is weak relative to the noise amplitude and lacks distance information. Single-frame target detection is unable to obtain the desired detection probability and false alarm rate because there are short of information about shape, size, texture and other features of targets. Because of target’s regular and continuous movement, multi-frame target detection in image sequences which employing temporal information to confirm real targets becomes an effective solution.To improve detection ability of IRSS in available devices, in-depth and comprehensive research work has been done on dim and small target detection and tracking in an imager based on staring Infrared focal plane array (IRFPA) in this dissertation. Small target’s imaging process in staring IRFPA is discussed, and then the mathematical models are presented. The dim and small target’s statistical characters including its size, shape, gray level, contrast and local correlativity between target and its ground are analyzed. Based on works above, a series of algorithms are present to detect dim and small targets based on scanning staring IRFPA, include: algorithm of Infrared image preprocessing based on finite adaptive neighborhoods to enhance dim targets and restrain noise, algorithm of background prediction based on mathematical morphology, algorithm of line-target constant false alarm ratio (CFAR) detection. To reduce further false alarm rate in single-frame target detection, a target re-affirm technique based on target local intensity distribution function matching is presented. Detection method based on multi-frame tracking is another advanced approach to high detective probability in clutter environment. After analyzing current track before detect (TBD) techniques, an optimized dim target detecting technique based dynamic programming is presented. The designed algorithm put forward a novel stage benefit function and publishment function based on target’s inter-frame relativity and consistency. This paper designs a multiple hypotheses tracking scheme for low altitude IRSS. The presented multitarget tracking (MTT) method adopts Kalman filter to predict new position of target, and a complete algorithm framework is brought forward for track intiation, tarck maintaining and track ending. In track maintaining, target’s position, mean gray level, size and local SNR are taken into count, and a new data association technique based on multiple target characters is investigated. An emulator of dim and small targets detection and tracking is designed to confirm validity of methods. Results show that algorihms of target detection and tracking in IR image sequences have high clutter suppression ability, anti-target-missing ability and significant real-time processing ability.Based on results of theoretic analysis and algorithm simulation above, an efficient and real-time infrared digital signal processing model for low altitude IRSS is developed. Testing shows that algorithms in this dissertation are efficient and practical.更多还原