图象目标识别技术的研究与应用

【作者】 舒海燕；

【导师】 郭雷；

【作者基本信息】 西北工业大学 ， 模式识别与智能系统， 2002， 硕士

【摘要】 目标识别在当今军事国防以及民用领域的作用日益显著，从航拍及遥感图象中实现目标的自动识别在近些年来取得了长足的发展，但是仍然没有一种可以普遍适用的方法。这其中的困难不仅体现在目标场景的复杂性本身，同时也是由于对人类视觉机制认识上的不足。 目前常用的目标识别算法有两种：一种是传统的由下而上的数据驱动型，另一种是由上而下的知识(假设)驱动型。前者常用于飞机、工件等团块目标的识别中；而后者是在复杂场景中将桥梁、道路、机场跑道等目标识别出来最为常用的识别方法。文章中对后一种方法做了较为详细的研究。 知识驱动的目标识别方法的实现有很多种，其中图象分割中所涉及到的区域分割和线状特征提取技术是我们的研究的基础，在文章中对这两种方法中的所涉及的具体知识做了详细的介绍。 知识的表示与利用在我们在自上而下的目标识别方式中起着十分关键的作用。文章中对知识的概念以及知识的表示方法给出了详细的介绍，并结合对国内外相关文献的研究，给出了一套目标识别的思路与流程。 在前面工作的基础上，文章针对常见的人工目标——桥梁、机场跑道，给出了实际的实现算法。机场跑道的实现算法主要结合了线状特征提取的思路，桥梁识别的算法则主要结合了区域特征分析的方法。这两种方法在结合前人研究的基础上，均有自己的独到之处。通过大量实验验证，算法的稳定性和实时性都非常好。其中根据机场跑道的识别算法文中还设计了一个简单的仿真目标跟踪系统。更多还原

【Abstract】 Target recognition is valued in military affairs. Although the researches in automated target detection from aerial and remote sensing imagery have made great progress in recent years, there is still no commonly accepted method. The difficulties come not only from the complexity of scene itself, but also from the weak knowledge about the principle of human visual perception.There are two basic target recognition algorithms. One of them is processed from bottom to top, which is called data-driving method. This traditional method is most often used in object as airplane, workpiece, etc. The other is processed from top to bottom, which is called knowledge-driving method. This one is always used in extracting road, bridge and runway from complex scene. This paper presents adequate research on the later one.Image segmentation techniques involving region segmentation and line feature extraction which are the foundation of our research are introduced. And the other concerning methods have been referred and exemplified.It has been well recognized that the expression and the application of knowledge play a key role in target recognition. The definition of the knowledge and several expression methods of the knowledge are introduced. On the basis of it, this paper gives a set of flow on how to extract target from complex scene based on the knowledge.At the final of this paper, two algorithms on man-made object recognition are present. Algorithm on runway recognition is based on linear feature extraction and the algorithm on bridge recognition is focused on region character analysis. All of them are robust and realtime. And a simulation tracking system is designed based on runway recognition algorithm.更多还原