图像目标识别跟踪及其实现

2019-02-24 21:44:08

算法 algorithm tracking 跟踪 目标









中文题名图像目标识别跟踪及其实现

 





副题名 





外文题名 





论文作者任世宏   





导师毛二可院士  何佩琨教授   





学科专业信号与信息处理   





研究领域\研究方向 





学位级别博士 





学位授予单位北京理工大学   





学位授予日期2001   





论文页码总数105页   





关键词制导  成像制导  跟踪  模式识别  实时图像处理   





馆藏号BSLW

/2001

/TJ765

/16 





【中文摘要】

    随着科学技术的的不断进步,现代制导武器朝着精确化、智能化的方向发展。“打了就不管”的精确制导技术已经成为现代制导武器的发展方向。这一潮流从算法和硬件平台两个方面对复杂背景下目标识别技术提出了更高的要求,本文的工作在算法理论分析及计算机模拟、系统设计和实现几个方面进行了以下研究:
   本文第一部分对复杂背景下目标识别算法涉及到的多个方面进行了较为深入的研究,并且将结果进行了计算机模拟和实际验证。首先介绍了几种计算灰度表面分形维数的算法。通过比较得出了Variation算法性能最佳的结论。提出了利用计算物体轮廓分形维数的方法区分人工目标和自然物体的快速识别方法,给出了通过链码的变换得到待识别目标的轮廓粗糙性曲线的具体算法。其次,在复杂背景中为克服噪声对目标几何和形状特征的影响,提出了一种以主观判断为准则,基于链码的边缘平滑算法,该算法可以和轮廓跟踪同时进行,能够在很大程度上减小背景及噪声对目标形状特征影响,同时满足实时性要求。再其次,在对各种图像模板匹配算法及图像特征匹配算法分析、比较的基础上,得出新算法是目前最好的一种图像配准算法的结论。通过对新图像特征配准算法的算法理论及运算量进行分析,给出了该算法的简化算法。同时为克服复杂背景照度不定带来的影响,提出了补偿措施,这样简化后的算法既保持了原算法固有的抗照度变化特性,又使得其运算量大大降低。
   本文第二部分设计了一个全程成像制导自动识别跟踪方案,即将制导过程分为三个阶段:初段、中段和末段。当目标较小或隐藏于背景之中,即制导初段时,采用匹配跟踪方式;当目标较大或特征较明显,即处于制导中段时,采用特征跟踪方式;当导弹临近目标或目标充满视场时,即制导末段时,系统又回到匹配跟踪方式。最后完成了整个过程的计算机模拟。
   最后在基于TMS320C80的实时真彩色图像处理系统上,实现了初步具备智能化面向工程应用的复杂背景下目标自动识别与跟踪系统,在对空目标的跟踪实验中取得了较好的跟踪效果,为实现真正“打了就不管”的成像制导技术打下了一定的基础。











【外文摘要】

 Abstract
   With the continuous development of science and technology, the modern guided weapons have been made more and more accurate and intellectual. The accurate guidance technology of"Launch and Forget"has already become the trend of the development of the modern guided weapons. This, in terms of both algorithm and hardware platform, has made still greater demands for the target-recognition technique in a complicated background. The thesis paper focuses on the theoretical analysis of algorithm, computer simulation, systematical design and its realization. The contents are as follows:
   First of all, the paper has made a thorough study of the various aspects concerning the algorithm of target-recognition in a complicated background. The conclusions have been tested and simulated with computer. At the beginning, the article introduces several algorithms of pixel surface fractal dimension. After comparison, it has come to a conclusion that variation algorithm is the best. It has provided not only a way of quickly distinguishing between an artificial object and a natural object, but also a way of getting the rough contour curve through the changes in chain code. Then, in order to reduce the influence of the noise on the geometric and formal features of an object, the paper suggests an algorithm based both on subjective judgement and on the contour smooth of chain code, which can be made in the process of contour tracking. And, may, to a great extent, reduce the influence of the background and noise on the formal features of the object, and, at the same time, meet the real time needs. What's more, based on the comparison and analysis of the algorithm of various mould matching and feature matching, it is concluded that the new algorithm is the best for image matching at present. Through the analysis of the algorithm theory of the new image feature matching, it has got a simplified algorithm. Also, to reduce the influence brought about by the unstable intensity of the lighting in a complicated background, it recommends some compensatory measures. Thus, the simplified algorithm not only maintain the ability to reduce the changes in the intensity of lighting, but also, greatly reduce the amount of algorithm.
   In addition, the writer of the thesis also designs a way of automatic recognition tracking guidance in the formation of an image, which divides the guiding process into three phases: the primary phase, the middle phase and the final phase. When the target is relatively small or hidden in the background, which is considered the primary phase, a matching tracking will be used. When the target becomes bigger or its features are more obvious, which is the middle phase, a feature tracking will be used. And when the guided missile comes nearer to the targets or the targets are full in the view field, which means coming into the final phase, the system, again, will use matching tracking, thus completing the whole process of the computer simulation.
   At last, a system of target automatic recognition and tracking has been realized, which is based on the TMS320C80 image processing system of real time and real color, It has primary intelligence and can be used in engineering.