基于矩不变量和神经网络识别三维物体  

Three-dimensional object recognition based on moment invariants and neural network

作　　者：苑惠娟[1] 孙樵[1] 尹双[1] 于晓洋[1] 

机构地区：[1]哈尔滨理工大学测控技术与通信工程学院黑龙江省高校测控技术与设备重点实验室,哈尔滨150080

出　　处：《黑龙江大学自然科学学报》2011年第2期267-270,共4页Journal of Natural Science of Heilongjiang University

基　　金：国家自然科学基金资助项目(61071051);黑龙江省自然科学基金资助项目(F201020);黑龙江省教育厅计划项目(11551102);哈尔滨理工大学优秀学科带头人支持计划项目

摘　　要：三维物体识别是计算机视觉的重要研究课题,广泛应用于工业和军事等领域。目前大多数识别方法,只能由某个特定角度的二维图像识别三维物体。采用矩不变量与神经网络相结合的方法实现三维物体的识别。使样本物体绕z轴旋转,间隔一定角度拍摄一幅图像(最大间隔角度不超过30度),得到一系列样本物体图像。计算这些图像的矩不变量并作为其特征向量,并且应用LVQ神经网络进行训练,从而实现三维物体的识别。实验结果表明,这种方法能够从不同角度的图像识别三维物体。Three-dimensional object recognition is an important research topic in computer vision,and widely used in industrial and military fields.In most method of recognition currently,3D object can be recognized only by 2D image of a particular point of view.Moment invariants and neural network are used to achieve 3D objects recognition.Rotating the sample object around the z axis,an image is captured when an angle turned（not more than 30 degrees）.So a series of sample images are obtained.3D objects recognition is realized by calculating the moment invariants of these images as eigenvectors,then training by using LVQ neural network.The result shows that this method can identify 3D objects by using image captured from different angles of the object.

关 键 词：三维物体识别 矩不变量 LVQ神经网络 特征向量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]