红外多目标实时跟踪方法的研究

	红外多目标实时跟踪方法的研究
	穆治亚
学位类型	博士
导师	魏仲慧
	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	靶场是对新兴武器装备性能进行试验检验和技术指标测试、鉴定而专门建造的设施齐备的广大地域，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。红外多目标跟踪技术，就是根据红外图像不同帧之间的差异性，对各个目标进行数据分析，得到各个目标不同的观测集合或轨迹，并通过分析目标的位置、速度、加速度、灰度特性等信息，预测并准确捕获其在下一帧图像上的运动位置及运动状态，实现对目标弹体及武器的试验检验过程。靶场测量中的多目标跟踪技术具有以下几个难题：首先，图像信噪比低、对比度差，目标尺寸小、形状和纹理等信息匮乏，图像背景复杂，目标与背景交织在一起，目标通常难以辨别，造成红外多目标跟踪十分困难。其次，目标交叉运动过程中的数据关联问题，目标的丢失与重现现象，也是多目标跟踪的难点之一。目前国内外的研究重点都集中在红外单目标的跟踪上，而红外多目标的跟踪领域却研究较少，因此，开展红外多目标实时跟踪方面的研究具有重要的学术意义和实用价值。结合当前工程实际需求，本文主要针对靶场测量中数量在10个以上的多批次连发目标群的实时跟踪问题展开研究，开展的研究工作主要有：1、对基于数字图像处理技术的图像预处理与多目标实时跟踪算法进行了讨论研究，为实现整个系统功能提供理论基础； 2、针对多批次连发目标数量较多、目标运动随机性较大、目标跟踪过程数据量大、跟踪过程必须实时等特点，提出了一种新的多站组合跟踪方法；3、靶场测量中的多连发目标跟踪过程为实时跟踪过程，因此需要对数据进行高速实时处理，本文对数字图像处理的硬件平台进行设计及研究。本文首先对面阵红外探测器的成像特性进行分析，研究讨论了红外图像中各个结构层次的特性，针对这种特性的差异性，采用基于稀疏编码理论的图像预处理算法，建立各个结构层次的超完备字典，可以针对不同的结构层次实现相应的预处理算法，使图像处理结果更加稀疏、简洁。其次，分析了单目标与多目标跟踪的原理，提出了一种采用马尔科夫（Markov）跳变系统与核聚类相结合的红外多目标跟踪算法，通过构建Markov跳变非线性系统的多目标转移模型，自回归位移预测模型（ARMA）对多目标跟踪过程中数据关联分析，解决了多目标跟踪问题中的目标间交汇、遮挡等问题，利用核聚类算法对粒子进行优化采样来提高粒子的贡献率，解决粒子滤波过程中的退化问题。再次，针对多批次连发目标数量较多以及目标运动随机性较大等特点，提出了一种多站组合跟踪方案，通过分析多批次连发目标群的运动特性，定义了三种不同类目标群的概念，提出了基于多尺度多级FCM和智能群决策的目标群分类决策策略，将目标群智能分类为不同的类目标群，利用图像融合决策器对多台经纬仪进行决策控制，实现不同类目标群的实时跟踪。最后，根据图像融合决策器的功能及多目标跟踪的技术要求，设计FPGA+双多核DSP处理器架构的图像融合决策器平台，通过多核处理器的并行数据处理、任务分配与信息交互提升系统对红外图像数据的处理速度。最后将提出的预处理算法、多目标跟踪算法及目标群分类决策策略进行了硬件移植，并通过外场多项试验测试，验证了算法和硬件平台系统方案的正确性、实时性以及可靠性。
其他摘要	Range is a vast and well-equipped place that is specially built for testing and detecting the performance of new weapons. What is more, range is also used to test and identify the technical indicators. In the range, we reveal the intrinsic and inherent performance characteristics of tested weaponry by measuring the flight trajectory parameters of target missile, which is the most direct and effective means. Infrared multi-target tracking technology is that based on the difference between different frames of the infrared image , we analyze the data of each target to gain different observation set or track, then through the analysis of the target position, velocity, acceleration, grayscale characteristics and other information, we accurately predict and capture the movement location and motion status of the next frame, finally we succeed in testing the experiments process of target projectile or weapon. However, multi-target tracking technology of measuring range meets the following challenges: on one hand, infrared multi-target tracking becomes very difficult because of low SNR, poor contrast ratio, small size of target, lack of information about shape and texture , complex image background and the mixture of objectives and background that make identification very difficult. On the other hand, data association problem coming from target consolidation and cross-motion process and the phenomenon of target lose and reappearance make multi-target tracking technology get into trouble. Recently, scholars who come from home or abroad do not pay more attention to multi-target tracking research instead of being concentrated on single-target tracking .So infrared multi-target tracking study plays an important role in academic region and has practical value. Combined with the actual needs of the current project, major research work includes the following points: 1, having a discussion on the image pre-processing and multi-target tracking algorithm based on digital image processing technology in order to provide theoretical basis for whole system function; 2, working out new tracking strategy using several Theodolites ,for the large number of multiple and continuous targets , great randomness of characteristics of target motion;3, designing and studying the hardware platform of digital image processing so as to address the multi-target tracking problem in measurement of the range . The paper analyzes the imaging characteristics of array infrared detector then study and discuss the characteristic of each structure of infrared images. According to the characteristic difference, we establish ultra-complete dictionary that help to come up with corresponding pre-processing algorithm being aimed at different level of structure to make the effect of image process more sparse and simple by adopting image preprocessing algorithm on the basis of sparse coding theory. At the same time, the article proposes infrared multi-target tracking algorithm that is the combination of MJNSs and kernel clustering. The algorithm can solve the multi-target tracking problems such as: target intersection, obstructions and so on through building the MJNSs model and ARMA model to analyze the related data generating from multi-target tracking process. In addition, by using kernel clustering algorithm, we optimize the sampling to improve the high contribution rate of the particles. And we also propose a several Theodolites tracking program in accordance with the great number of continuous shoot and high randomness of target motion. By analyzing the motion characteristics of the target group, we give different conception for three various target groups and figure out classification and decision-making strategy of target group based on Intelligent Decision Control and multi-scale and multi-level FCM. The strategy classifies automatically target group into different class target groups. We use Image Fusion Decider to control several theodolites ,which is to track different target groups. We build Image Fusion Decider platform that consists of FPGA processor and dual multi-core DSP processors according to the function of Image Fusion Decider and the requirement of multi-target tracking technology then raise the data processing speed of infrared image with parallel data processing, task allocation and information exchange of multi-core processors. Finally, we make hardware transplant based on the proposed pre-processing algorithm, multi-target tracking algorithm and classification and decision-making strategies of target group. At the same time we verify the correctness and feasibility of the algorithm and system solutions of hardware platform by testing a great deal of actual range .
语种	中文
文献类型	学位论文
条目标识符	http://ir.ciomp.ac.cn/handle/181722/41452
专题	中科院长春光机所知识产出
推荐引用方式 GB/T 7714	穆治亚. 红外多目标实时跟踪方法的研究[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
穆治亚.pdf（2656KB）			开放获取	CC BY-NC-SA	请求全文