一种基于支持向量机的三维物体识别方法J.docVIP

一种基于支持向量机的三维物体识别方法* 徐 胜 彭启琮 管 庆 赵明渊 (电子科技大学通信与信息工程学院, 成都 610054) 摘 要: 提出从三维物体的二维图像中提取仿射不变傅氏描述子、色彩矩及纹理特征, 组成一个25维的特征向量, 送入支持向量机训练并用于三维物体识别。算法利用了仿射不变傅氏描述子在物体发生仿射形变时具有不变性, 利用色彩矩和纹理特征区分形状相似但有不同颜色及纹理的物体, 并引入支持向量机作为分类器。基于三维物体图像数据库COIL-100测试了算法的识别性能, 当每个物体训练样本图像数量为36个(视角间隔10°)时达到了100％的识别率, 进一步减少训练视角数量也达到较满意的识别性能。 关键词: 三维物体识别；仿射不变傅氏描述子；色彩矩；纹理分析；支持向量机 中图分类号: TP391.4　　　文献标识码: A　　　国家标准学科分类代码: 520.2060 A method for 3-d object recognition based on support vector machine Xu Sheng Peng Qicong Guan Qing Zhao Mingyuan (School of Communication and Information Engineering, Univ. of Electron. Sci. Tech. of China, Chengdu 610054, China) Abstract: To improve the performance of 3D objects recognition system, a method is proposed to extract the affine-invariant Fourier descriptors (AIFD), color moments and texture features from the 2D images of 3D objects. These features are combined into a feature vector of 25 components and sent to support vector machines for training and recognition. The invariance properties of AIFD under affine transformation (translation, scaling, rotation and shearing) for an object are used in this method. And color moment and texture features are used to distinguish the object of similar shape and different appearance. Support vector machines are used as classifier. Based on the public 3D objects dataset COIL-100 the method was assessed. 100% correct rate of recognition is achieved when the number of presented training views for each object is 36 (10 degrees interval). When the number of training views is reduced, the correct rate of recognition is also satisfied. Keywords: 3D object recognition; AIFD; color moments; texture analysis; support vector machine 1 引 言 目前出于城市规划、电子交通监控、工业测量自动化、军事侦察及医疗仪器等各个领域的大量应用需求, 三维物体识别已成为一个较活跃的研究领域, 具有很大的实用价值和重要意义[1]。基于外观(appearance-based)或基于视图(view-based)的三维物体识别算法研究, 近来成为人们研究的热点[2-5]。虽然基于视图的方法不易克服物体重叠问题, 但当物体的几何建模很困难或根本不可能得到时, 就可利用该方法来识别三维物体[6]。在一副二维图像中, 三维物体的外观取决于其形状, 反射特性, 姿态和环境亮度等, Murase等较早的提出了参数化的特征空间方法, 直接从三维物