SIA OpenIR  > 光电信息技术研究室
面向成像制导的目标跟踪方法研究
Alternative TitleResearch on Target Tracking for Imaging Guidance
陈宏宇1,2
Department光电信息技术研究室
Thesis Advisor罗海波
Keyword成像制导 可变形模型 相关滤波 重检测 长时目标跟踪
Pages121页
Degree Discipline模式识别与智能系统
Degree Name博士
2020-11-26
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract精确制导武器因其命中精度高、打击距离远、附带损伤小等特点,在现代战争中发挥着越来越重要的作用。作为一种重要的末制导方式,成像制导是其走向精确化和智能化的有效途径。目标跟踪算法作为成像制导系统中非常重要的一个环节,决定着成像制导武器的打击效果。现存的目标跟踪方法在简单场景下能够完成跟踪任务,然而在复杂的战场环境中,制导武器会受到环境及打击目标的影响导致命中率下降,如目标发生形变、光照变化、复杂背景干扰、目标遮挡、尺度变化等。因此,设计出精度高、鲁棒性好能够应对多种挑战的目标跟踪方法是一项极具挑战性的任务。本文以成像制导为应用背景,对成像制导中目标与背景的复杂性进行了分析,对现有的目标跟踪方法进行了综述并分析了其存在不足。论文主要对跟踪过程中遇到的目标形变、尺度变化、遮挡等问题进行了研究,展开了面向成像制导的目标跟踪方法研究。主要贡献总结如下:1.分析了基于可变形模型对目标进行表征的优越性。针对成像制导应用无法实现自动子块提取的问题,提出了一种基于多特征融合的子块自动提取方法。该方法首先采用基于人眼视觉注意机制的显著性检测方法对子块的区分性进行表征;其次,利用边缘方向离散度对子块的纹理丰富度进行度量;然后,融合上述特征获得联合适配性置信度,并根据目标的面积和宽高比自适应确定子块的个数和尺寸;最后根据联合适配性置信度提取目标子块。为基于可变形模型的目标跟踪算法的工程化应用奠定了基础。2.提出了一种可变形模型与相关滤波相结合的目标跟踪方法。为了充分利用相关滤波算法精度高、效率高的优点,本文将其与可变形模型相结合,既能够解决目标的部分遮挡及形变问题,又能实现对目标的精确稳定跟踪。在预测目标的尺度时,采用序贯蒙特卡洛框架根节点和子块的尺度进行分别估计,更加符合形变目标的特性。另外,根据子块的置信度,自适应地获取联合置信度图,完成目标位置的精确估计。同时,将所提出的跟踪方法在公开数据集以及自采数据集上进行测试,实验结果表明,该方法能够解决尺度变化、目标形变及部分遮挡的问题。3.提出了一种长时目标跟踪框架。跟踪过程中,目标可能受到云层、人工干扰物的遮挡,为了解决目标受到严重遮挡后的重新捕获问题,提出了一种基于子块更新的长时目标跟踪方法,该方法能够在目标处于视野内时稳定跟踪目标,当目标发生不可信跟踪时,根据自适应阈值启动重检测模块。在精确定位目标位置与尺度估计的同时,具备较强的抗干扰能力。实验结果表明,该方法能够有效应对目标遮挡及消失再现的准确跟踪问题。4.设计了面向成像末制导的目标跟踪流程。根据弹目距离的变化,分析了目标成像特性,提出了跟踪方法选择策略。针对空对地和空对空两种应用场景,设计了两种跟踪流程并进行了详细分析,完成了将论文研究成果向实际装备推广的初步探讨。
Other AbstractPrecision-guided weapons are becoming more and more critical in the modern war due to their high hitting accuracy, long striking distance, minor collateral damage, etc. studied by all countries in the world at present. As a significant terminal guidance method, imaging guidance is an effective way to make it precise and intelligent. In the imaging guidance system, target tracking algorithm is a vital component that determines the impact effect of the imaging guidance weapon. As a vital part of the imaging guidance system, the target tracking algorithm determines the impact effect of the imaging guidance weapon. Most existing target tracking algorithms could track the target in simple scenes. However, the hit accuracy of the guided weapon might be affected by a complex battlefield environment, or he attacked target, such as scale variation, deformation, occlusion, etc. Thus, it is a challenging task to design a target tracking algorithm that can deal with multiple tracking challenges and achieve high precision and robustness. In this thesis, the target and background in imaging guidance have been analyzed. The existing target tracking algorithms are reviewed, and the shortcomings of the current target tracking algorithms are analyzed. This thesis mainly aims at tracking challenges in imaging guidance, including deformation, scale change, and occlusion. The main contributions of this thesis are summarized as follows: 1.This thesis analyzes the superiorities of the deformable part model in representing the target. To solve the problem that tracking methods based on deformable parts model cannot select parts manually in the imaging guidance application, an automatic parts selection method based on multi-feature fusion is proposed. Firstly, a saliency measure based on the visual attention model is used to measure the local contrast of parts. Secondly, edge direction dispersion has been employed to describe the richness of texture details. After obtaining the joint suitable-matching confidence map, the number and size of parts are adaptively selected according to the pixel area and aspect ratio of the target. This automatic parts selection method lays a foundation for the engineering implementation of target tracking method based on deformable part model. 2. A target tracking method based on deformable part model and correlation filter is proposed. Due to the high precision and efficiency of the correlation filter, our tracking method combines it with a deformable part model, which can solve the problem of partial occlusion and deformation of the target and achieve accurate and robust tracking. To achieve more accurate scale estiamation, the sequential Monte Carlo framework is used to estimate the root and parts, respectively. In addition, according to the confidence score of parts, the joint confidence map is obtained adaptively to accurately estimate the target position. At the same time, the proposed tracking method is evaluated on public data sets and the sequences obtained by us. Experimental results show that our target tracking method could solve tracking challenges, including scale variation, deformation, and partial occlusion. 3. A long-term target tracking framework is proposed. During tracking, the guaidance weapons are faced with the challenges caused by clouds and artificial jamming. To re-capture the lost target under full occlusion, this thesis proposes a long-term target tracking method based on reliable parts. This method could track the target precisely when the target is in the field of view or is occluded partially. When unreliable tracking result is detected, the re-detection module will be activated according to the adaptive threshold. When locating the position and estimating the scale of the target, our method shows the excellent anti-jamming capacity. Experimental results show that the tracking method proposed in this thesis could effectively deal with full occlusion and out-of-view tracking challenges. 4. The tracking workflows for imaging terminal guidance is designed. According to different missile-target distance, the appearance of target has been analyzed and a tracking method selection strategy has been proposed. Aiming at the application scenarios of air-to-surface and air-to-air, two tracking workflows have been designed and introduced in detail. A preliminary discussion on the application of the research results of the thesis generalize to the actual equipment has been done.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/27975
Collection光电信息技术研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院大学
Recommended Citation
GB/T 7714
陈宏宇. 面向成像制导的目标跟踪方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2020.
Files in This Item:
File Name/Size DocType Version Access License
面向成像制导的目标跟踪方法研究.pdf(7066KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[陈宏宇]'s Articles
Baidu academic
Similar articles in Baidu academic
[陈宏宇]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[陈宏宇]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.