【作者】 赵军；

【导师】 曲仕茹；

【作者基本信息】 西北工业大学 ， 机械设计及理论， 2004， 硕士

【摘要】 本文的研究得到西北工业大学研究生创业种子基金项目资助(Z20030060)。 世界局势的局部动荡和经济与科技的发展使得飞机识别在当今国防军事以及民用领域的作用日益显著。从航拍及遥感图像中实现飞机的自动识别在近些年来取得了长足的发展，但仍有许多方面需要作更进一步的研究，其中的困难不仅表现在目标的复杂多样性，也有对人类视觉机制认识不足的原因。 目前对三维物体识别的研究方法大体可分为基于视觉计算理论、基于感知组织理论、主动视觉以及基于模型的视觉等类型。本文对飞机识别的研究应用了基于模型的视觉方法——将CAD模型转换为适合视觉应用的二维模型，通过目标的单幅灰度图像与二维模型的识别匹配达到对三维目标识别的目的。 建模型的过程中，首先运用逆向工程测量方法，获得曲面物体上不同截面轮廓线的测量数据点列，接着对测量数据点进行平滑处理，用最小二乘法求解基于数据点列的控制多边形，以非均匀三次B样条曲线拟合截面轮廓曲线；然后利用基于垂距的数据采样算法，对拟合曲线上的数据点进行重新采样，达到优化数据点的分布及减少描述曲线数据量的目的。将重新采样得到的曲线特征点列写成CAD造型软件Pro／E的接口文件格式，在Pro／E完成物体外形轮廓曲面重构和建模。仿真结果表明算法简便、有效。对飞机的识别还提出了一种利用仿射不变量——同底三角形的面积比，进行自动识别的新算法。该方法用多边形近似代替已提取出的飞机边缘轮廓，通过对该特征多边形的识别从而达到对飞机的识别。在对多边形进行识别时，用描述子即多边形的顶点个数以及以最长线段为底的同底三角形面积比向量，来描述多边形的形状。利用建立的三维CAD模型生成全方位姿态模型库来克服飞机姿态不同时飞机部件相互遮挡所造成的图像差异。进而建立姿态匹配特征模板库，通过搜索姿态模板库匹配多边形描述子来识别目标。整个识别过程逐层递进，由简到繁，从而避免了大量的冗余操作。实验表明该算法是高效可行的。更多还原

【Abstract】 Today, aircraft recognizing become more important in field of national defence and civil aviation due to the development of economy and technology and local confliction. It has gained a great deal of achievement in identifying aircraft automatically from airphoto and remote sensing image. But the deeper study is necessary in many fields. The reason is not only the variety of complex object, but also the insufficiency of knowledge to human vision mechanism.Presently, there are three methods of 3D object recognition: based on the theory of vision calculate, based on initiative vision and based on model. In this paper,recognition aircraft is based on model--it is realized through matching the singleimage with 2D model.A new algorithm about adaptive data sampling based on vertical distance has been proposed, furthermore, a modeling method applying this algorithm for complex surface objects has been addressed in this paper. This process is as follows: firstly, the data of different section contour and of section line have been obtained by reverse-engineering; Secondly, Non-Uniform B-Spline approximation algorithm is used to fit the discrete data; Then the data have been optimized and reduced by adopting adaptive sampling of key points of the fitted curve based on vertical distance, sequently the adaptive sampling data is transformed into the format of the .ibl file of the famous 3-D design software Pro/E. therefore we reconstruct the surface and a model is generated; Lastly, The effectiveness of the adopted algorithm and modeling approach are demonstrated by the experiments.In part of recognition, two problems are studied. First, shape descriptors, which are invariant under affine transformation, are proposed to describe the polygon. Second, a new algorithm for recognizing aircrafts from a single image is presented. The polygon is employed to present the aircraft contour approximately, and the aircraft can be recognized by recognizing the characteristic polygon. Then the shape descriptors are introduced to describe the polygon. The shape descriptors are thenumber of the polygon’s vertices, and the triangles areas ratio vector of the longest line hemline. The database of the aircraft’s each pose is established to overcome the image differences which are produced by the occlusion of the aircraft’s components in different pose. The aircraft is recognized by search for the database of the aircraft’s each pose. The recognition method is a process from simple to complex, thus many redundancies calculation are avoided. The experiment results show the algorithm is efficient and feasible.This thesis is supported by Graduation Seed Fund of Northwestern PoIytechnicalUniversity(Z20030060)更多还原