【作者】 罗诗途；

【导师】 罗飞路；

【作者基本信息】 国防科学技术大学 ， 仪器科学与技术， 2005， 博士

【摘要】 在未来高技术条件下的局部战争中,装甲战车乘员要在对目标作战的瞬间处理大量的信息,这就需要一个将传感器、处理机和显示器等装置结合在一起的目标自动跟踪系统。这种系统应能从复杂的和混乱的散射干扰背景中迅速、可靠地提取目标,从而使乘员能尽快地对敌目标作出反应。但由于装甲战车的工作环境恶劣、作战环境多样化、目标种类繁多,而且跟踪系统对其所处平台的振动干扰较为敏感,这些都导致了研制装甲战车图像跟踪系统存在着较大的难度。本文探讨了图像跟踪系统用于装甲战车所存在的技术难点,研究了车载环境下目标的检测识别与跟踪技术,并进一步研究了装甲战车图像跟踪系统的实现技术,完成了相应工程项目的研制任务。本文的研究工作包括以下几个部分:①建立了装甲战车图像跟踪系统的跟踪理论。首先,根据对系统任务及战术技术指标的分析,建立了系统的数学模型并给出了系统的跟踪方程;然后,通过探讨如何求解该跟踪方程,将系统进行状态划分,并根据各个工作状态的描述,建立了系统的基本功能结构。论文通过分析系统的基本理论,指出系统的难点和本课题研究的重点。②针对车体振动对图像序列带来的干扰问题,提出了一种基于特征匹配的实时电子稳像算法。首先,对电子稳像中关键环节——图像配准技术进行了深入研究,给出一种由粗到精、由局部到全局的高效配准算法;接着,采用了递归Kalman滤波技术对配准参数进行运动滤波并对实际图像进行运动补偿;最后,利用了图像拼接技术进行“无定义区”重建,避免了普遍存在的信息丢失和图像降质问题。该算法在保证实时性的同时,具有很高的稳像精度。③为了使系统在简单场景下能够通过自动选取分割算法来提取目标,提出了一种基于粗糙集理论的图像分割智能决策方法。首先选取若干具代表性的分割算法构成算法库,并用它们对各种样本图像进行分割;然后利用从样本图像中提取出来的各种数值特征,并根据图像分割质量评价标准评判出各样本图像的最优分割算法,用其构成决策信息表;最后应用粗糙集理论来对决策信息表进行离散化处理和属性约简,以生成图像分割算法选取的决策规则。该决策方法能比较有效地根据系统所处理图像特征选取出算法库中最优的分割算法,并可满足系统的实时性要求。④提出了一种综合运用分形和特征匹配等技术,将复杂场景中目标提取出来的方法。首先,对传统的Snake模型进行改进,并将其应用到初始模板的建立中;然后,引入分形布朗随机场模型,利用分形维数和分形拟合误差确定可能的目标区域;最后,定义了一种新的最小失配距离(MMD)相似性度量,并基于目标的特征区域进行快速相关匹配,从可能区域中提取出目标。该算法通过精确建立初始模板和由粗到精的目标搜索策略,既保证了目标提取的精度、速度,又能对各种噪声干扰有较强的抑制。⑤对跟踪系统软、硬件平台的实现技术进行了研究。设计了以DSP为核心,结合PC104嵌入式计算机的双处理器并行处理硬件平台;并给出了嵌入式计算机和图像采集处理板的更多还原

【Abstract】 In the future local war of high technology, people on the armored vehicle must processabundant information in the moment of battling against the enemy target, so an automatic targettracking system is needed which hangs sensors, processor and display and so on together. Such asystem should extract the target from complicated and chaotic background of scatteringdisturbance quickly and reliably, thereby people on vehicle can react to the enemy target asquickly as possible. However, the working condition of armored vehicle is abominable andmultiplicate, the targets are diversified, and the tracking system is sensitive to the platform’s vibration. All these things have caused great difficulty in developing image tracking system onarmored vehicle.This paper probes into the technical difficulties in the application of image tracking systemon armored vehicle, and studies on the target detection, recognition and tracking techniques onvehicle. It also investigates the realizing technology of image tracking system on vehicle, andaccomplishes the development task of corresponding project.The primaryresearch of this paper includes:①Setting up the tracking theory of image tracking system on armored vehicle. First, thepaper constructs the system’s mathematical model and brings out the tracking equation according to the analysis of system task and tactics-technology index. Second, it divides the state of systemin order to solve the tracking equation, and constructs the basic function structure of systembased on the description of each working state. By analyzing the system’s basic theory, this paperpoints out the system difficultyand our study emphasis.②Putting forward a real-time electronic image stabilizing algorithm (EIS) in allusion to thedisturbance to image sequence by vehicle vibration. First, it delves into the key technique of EIS—image registration, and presents an effective registration algorithm which is from coarse toprecise and form local to global. Second, it applies recursive Kalman filtering technique inmotion filtering of registration parameters and makes motion compensation for practical image.Finally, it takes advantage of mosaicking to reconstruct undefined regions to avoid theubiquitous information missing and image degradation. This algorithm has high precision as wellas high speed.③Bringing forward an intelligent decision method of image segmentation based on roughset theory to make the system automatically select segmentation algorithm in simple scenes.Firstly, it selects some representative segmentation algorithms to make up of an algorithm library,which is used to process kinds of sample images; secondly, it makes the decision informationtable utilizing diversified numerical features extracted from the sample images and the optimalsegmentation algorithm of each sample image according to segmentation quality evaluationcriterion; finally, it applies rough set theory on discretization and attribution reduction of更多还原