序列图像红外小目标检测与跟踪算法研究

	序列图像红外小目标检测与跟踪算法研究
	孙继刚
学位类型	博士
导师	朱明
	2014-07
学位授予单位	中国科学院大学
学位专业	机械电子工程
摘要	红外成像技术具有抗干扰、隐蔽性好和适应天候能力强等优点，在军事领域和民用领域中的应用十分广泛。现代战争对武器装备和预警系统提出了更高的要求，因此如何尽可能提高目标的检测距离，更早更快的发现目标和跟踪目标是目前红外图像研究领域的一个热点和难点。因为成像距离远，成像面积小，目标在图像中多以小目标的形式存在，而且受到噪声和背景杂波的影响更大，这些条件增加了目标检测和跟踪的难度。如何在复杂背景条件下实现对红外弱小目标稳健的检测与跟踪具有重要的研究意义。本论文围绕红外弱小目标检测和跟踪的相关技术进行了深入的研究，下面介绍论文的研究工作和主要研究成果。在图像预处理方面，本文研究了背景抑制的基本原理和经典算法。针对传统双边滤波和加窗的双边滤波算法的局限性进行了分析，本文提出了一种基于Tophat变换和加窗的双边滤波相结合的图像预处理算法，实验证实这种算法有效的提高了系统的背景抑制能力和目标检测性能。在目标检测方面，本文详细研究了经典的阈值分割算法、尺度空间理论和DoG尺度空间算法，利用DoG算法完成了两项单帧检测实验，即指定目标尺度的目标检测和非指定目标尺度的目标检测。在DoG算法的基础上，提出了一种新的多帧检测算法：即DoG算法和管道滤波算法相结合的管径可自适应变化的管道滤波算法——PDAF算法。在目标跟踪方面，本文详细研究了卡尔曼滤波算法和粒子滤波算法。针对卡尔曼滤波算法对小尺寸目标难以精确跟踪的问题，本文提出一种卡尔曼滤波和DoG尺度空间算法相结合的新算法。用卡尔曼滤波器对目标运动位置实施初步预估，然后在跟踪门内应用DoG算法对目标进行检测，获得目标的精确位置和尺度参数，然后利用得出的参数值作为观测值对卡尔曼滤波器的参数进行更新，以利于卡尔曼滤波器准确预测下一帧小目标的位置，实验证明本算法可以有效提高卡尔曼滤波器定位的精确性。针对目标跟踪过程中目标尺度不断变化的问题，为了充分发挥粒子滤波在跟踪大尺寸目标时的优点，本文在尺度空间理论的基础上，设计了一种算法切换策略：利用DoG尺度空间算法提取目标尺度，当目标尺度在阈值以下时采用卡尔曼滤波算法对目标跟踪，当目标尺度在阈值以上时采用粒子滤波算法对目标进行跟踪。本文所提出的算法很好的结合了卡尔曼滤波和粒子滤波的各自优点，能够有效提高系统的跟踪性能。最后，为了满足目标跟踪的实时性要求，本文设计了一套基于DSP+FPGA的跟踪器硬件平台，验证了算法和硬件平台系统的可行性。
其他摘要	Infrared imaging technology has many advantages such as strong anti-interference ability, well-concealment and good weather adaptability, it gets a widely use in the military and civilian applications fields. Modern warfare put forward higher requirements to the weaponry and early warning systems, so how to improve the detection distance , earlier and faster finding the target and tracking the target is a hot and difficult research in the field of infrared images currently. Because of the long imaging distance, the target forms a small area in the image. By the great impact of noise and background clutter, these conditions increase the difficulty of target detection and tracking. How to achieve in complex background conditions has important significance for the study of robust infrared small target detection and tracking. This dissertation focuses on infrared small target detection and tracking technologies in-depth study and research. Here are the results of research work and major research papers. In the image preprocessing, this paper studied the basic principles and classical algorithms of background suppression. To overcome the limitations of traditional bilateral filtering and windowed bilateral filtering algorithm, this paper presents the image preprocessing algorithm based on the combination of Tophat transformation and windowed bilateral filtering. Experiments confirmed that this algorithm effectively improves the system background suppression and target detection performance. In terms of target detection, the paper detailed study of the classical threshold segmentation algorithm, scale space theory and DoG scale space algorithm. This paper completed two single-frame detection experiments using DoG algorithm, one is the specified scale target detection，the other is non-specified scale target detection .Base on the DoG algorithm, we proposed a new multi-frame detection algorithm: the combination of DoG algorithms and pipeline filtering algorithm. It is called diameter adaptive pipeline filtering——PDAF algorithm. On target tracking, this paper got a detailed study of the Kalman filter algorithm and particle filter algorithm. Kalman filter cannot accurately track the small targets, we propose a new algorithm that combines the advantages of Kalman filter and DoG algorithm. Using Kalman filter to imply the initial estimates of position of the moving target, and then use the DoG algorithm to get the exact location and scale parameters of the target in the tracking wave gate. We use the obtained parameter values as the observed values to update the parameters of the Kalman filter, which is conducive for Kalman filter to accurately predict the location of small targets in the next frame. Experiments show that the algorithm can effectively improve the positioning accuracy of the Kalman filter. There is a scale evolving issue in target tracking process. In order to give full play to the advantages of particle filter in tracking large targets, we design an algorithm switching strategy base on scale space theory: using the DoG algorithm to get the target scale, when the target scale is below the threshold using the Kalman filter algorithm for target tracking, when the target scale is above the threshold using particle filter algorithm. The proposed algorithm combines the respective advantages of Kalman filter and particle filter and it can effectively improve the tracking performance of the system. Finally, in order to meet the requirements of real-time target tracking, we designed a DSP + FPGA-based hardware platform tracker and validated algorithms and hardware platform.
语种	中文
文献类型	学位论文
条目标识符	http://ir.ciomp.ac.cn/handle/181722/41467
专题	中科院长春光机所知识产出
推荐引用方式 GB/T 7714	孙继刚. 序列图像红外小目标检测与跟踪算法研究[D]. 中国科学院大学,2014.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
孙继刚.pdf（2857KB）			开放获取	CC BY-NC-SA	请求全文