飞行器三维不变矩特征提取与识别研究

项目名称： 飞行器三维不变矩特征提取与识别研究

项目编号： No.61502389

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 其他

项目作者： 杨波

作者单位： 西北工业大学

项目金额： 21万元

中文摘要： 飞行器识别对于研发先进的飞行仿真系统以及培训试飞员的应变反应、专业技能具有重要意义。针对基于二维图像的飞行器识别方法在特征表示、区分能力上表现不足的情况，本项目研究一种基于三维图像和三维模型的飞行器特征提取和识别方法。该方法充分利用三维图像和模型在表示三维空间结构和三维几何外形上的优势构造飞行器的三维特征。考虑到飞行器识别任务中，特征描述方法应该对飞行器在任意姿态、任意位置、任意尺寸的情况下具有唯一性描述，本项目设计一种基于正交不变矩的特征描述方法，实现对飞行器平移、旋转、比例不变特征的提取。利用高阶不变矩、多尺度特征来构造飞行器的特征向量，结合主成分分析方法来建立互不相关的特征，进一步提取飞行器的三维整体与细节特征，实现高准确率的飞行器识别。本项目分别采用人造图像和实际拍摄的图像来测试评估飞行器识别效果。

中文关键词： 飞行器识别；Gaussian-Hermite；不变矩；三维不变矩特征；对称性；三维模型

英文摘要： Aircraft identification is important to design advanced flight simulator system, as well as to train the adaptability and professional skills of test pilots. Aircraft identification based on 2D images having the limitations to feature representation and discrimination, the proposal is focused on aircraft identification based on 3D images and models. The proposed method takes advantage of the abilities of 3D images and models in representing spatial structures and 3D geometric shapes to construct 3D features of aircraft. Feature descriptors should generally be invariants regardless of pose, position and size in recognizing 3D objects. Hence, a kind of feature descriptors based on orthogonal moment invariants are designed in the proposal. They enable to extract the features of aircraft which are independent of translation, rotation and scaling. Feature vector is constructed via high-order moment invariants and multiscale features. It is further refined by Principal Component Analysis (PCA) to achieve mutual independent features to deeply extract both 3D global and detailed features of aircraft and accomplish high accuracy aircraft identification. The proposed method is respectively tested by synthetic 3D images and real ones.

英文关键词： aircraft identification;Gaussian-Hermite invariants;3D moment invariant features;symmetry;3D models