【作者】 张焱；

【导师】 沈振康；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2008， 博士

【摘要】 地面背景下成像目标跟踪技术是对地打击精确制导武器成像导引头信息处理的重要组成部分。论文针对成像目标跟踪技术的应用需求,研究了地面背景下的运动单目标跟踪问题、运动多目标跟踪问题、和地面固定目标匹配跟踪问题。论文的主要工作如下:1、通过回顾现有的成像目标跟踪方法及其适应范围,分析了地面背景下成像目标跟踪技术的难点及创新需求。2、提出了一种结合广义交互式遗传算法的粒子滤波方法,改善了粒子滤波迭代过程中常见的粒子退化及匮乏问题。仿真实验说明该方法的滤波效果优于EKF滤波方法、传统的粒子滤波方法、和正则化粒子滤波方法。基于该方法实现了地面背景下的运动单目标跟踪处理;实际数据上的实验说明,该方法能有效克服强背景杂波以及目标被短时遮挡的情况。3、提出了一种基于视觉注意模型的运动多目标检测方法。该方法首先提取灰度、细节、运动等多种底层特征,然后融合这些特征形成动态显著性特征,最后利用注意焦点检测出同时存在的多个运动目标。实际数据上的实验验证了该方法较为稳健。4、提出了一种基于动态显著性特征的粒子滤波动目标跟踪方法,并将这一方法与显著性检测相结合,提出了一种地面背景下运动多目标跟踪方法。该方法将目标的动态显著性特征作为粒子滤波的状态向量,估计目标位置;同时与显著性检测得到的目标位置关联,完成航迹管理,实现多目标跟踪。实际数据实验验证了该方法能解决目标的出现、消失、合并、分裂以及被障碍物遮挡等难题。5、提出了基于梯度和最大值的相似性测度,和结合遗传算法与“高斯和”梯度神经网络的组合式搜索策略。在此基础上建立了一种新的地面固定目标匹配跟踪方法。可见光以及红外两组图像的仿真实验,验证了该方法具有较强的实用性及鲁棒性。更多还原

【Abstract】 Ground target tracking technique is an important part in the processing of imagery guiding heads in precise hitting weapons. Following the demands from practical applications of ground target tracking technique, this dissertation researches the problems of moving ground target tracking, ground multiple moving target tracking and matching-based ground fixed target tracking. The major parts of this paper are:First, the state-of-art target tracking methods for imagery sensors and their preliminary conditions are reviewed. Furthermore, the difficulties and demands in the research on ground target tracking are analyzed.Second, a novel Particle Filtering (PF) with Broad Interactive Genetic Algorithm (BIGA) is presented. The new method improves the performance on the problem of particle degeneration and particle shortage. Simulated experiments demonstrated that the presented method over-performs EKF, the traditional PF and the Regularized PF (RPF). Based on the presented method a ground moving target tracking procedure is implemented. Real-data experiments show that the procedure can cope with complex background clutter and short-time target occlusions.Thirdly, a novel moving multiple targets tracking method based on the visual attention model is proposed. The features on gray-level, detail and motion are extracted and then fused into a dynamic salient feature; moving multiple targets are detected with this dynamic salient feature based on the visual focus theory. Real-data experiments demonstrate that the proposed method has a robust performance.Fourthly, a dynamic salient PF (DSPF) target tracking method is presented. Combined DSPF with dynamic salient target detection, a robust ground multi-moving target tracking method is proposed. The new tracking method use dynamic salient features as the input to PF to estimate the target location; and associate with the dynamic salient target detection results to reveal the trajectories of the multiple moving targets in the view. Real-data experiments show that the proposed tracking method can cope with the target changes such as disappearing, merging, splitting and occulting.Fifthly, we proposed the maximum of gradient sum as the similarity measurement and a search strategy composed of the Genetic Algorithm (GA) and an improved "Gauss Sums" Gradient Neural Network (IG-GNN). Based on this similarity measurement and search strategy, a novel matching-based ground fixed target tracking method is presented. The experiments on synthetic optical and IR images show that the method is robust and ease to be implemented in practical applications.更多还原