电子学报 ›› 2015, Vol. 43 ›› Issue (11): 2277-2283.DOI: 10.3969/j.issn.0372-2112.2015.11.021

• 学术论文 • 上一篇    下一篇

一种基于2D和3DSIFT特征级融合的一般物体识别算法

李新德1, 刘苗苗1, 徐叶帆1, 雒超民2   

  1. 1. 东南大学自动化学院复杂工程测量与控制教育部重点实验室, 江苏 南京 210096;
    2. 底特律大学电子与计算机工程系, 美国 底特律 48221
  • 收稿日期:2015-01-08 修回日期:2015-06-02 出版日期:2015-11-25
    • 通讯作者:
    • 李新德
    • 作者简介:
    • 刘苗苗 女,1990年2月出生,河南安阳人.东南大学自动化学院硕士研究生.主要研究方向为模式识别与分类.E-mail:374561475@qq.com
    • 基金资助:
    • 国家自然科学基金 (No.60804063,No.61175091); 航空基金 (No.20140169002); 江苏省"青蓝工程"资助计划; 江苏省"六大高峰人才"资助计划

A Recognition Algorithm of Generic Objects Based on Feature-Level Fusion of 2D and 3D SIFT Descriptors

LI Xin-de1, LIU Miao-miao1, XU Ye-fan1, LUO Chao-min2   

  1. 1. Ministry of Education Key Laboratory of Measurement and Control, School of Automation, Southeast University, Nanjing, Jiangsu 210096, China;
    2. Department of Electrical and Computer Engineering, University of Detroit Mercy, Detroit, Michigan 48221, USA
  • Received:2015-01-08 Revised:2015-06-02 Online:2015-11-25 Published:2015-11-25

摘要:

如何选择合适的特征表示一般物体类间差异和类内共性至关重要,因此,本文在2D SIFT(Scale Invariant Feature Transform,SIFT)的基础上,提出了基于点云模型的3D SIFT特征描述子,进而提出一种基于2D和3D SIFT特征级融合的一般物体识别算法.分别提取物体2维图像和3维点云的2D和3D SIFT特征描述子,利用"词袋"(Bag of Words,BoW)模型得到物体特征向量,根据特征级融合将两个特征向量进行融合实现物体描述,运用有监督分类器支持向量机(Support Vector Machine,SVM)实现分类识别,给出最终识别结果.最后,实验验证了本文提出算法的好处.

关键词: 一般物体识别, 点云, 2D SIFT, 3D SIFT, 特征融合, BoW模型, SVM分类器

Abstract:

How to choose the appropriate feature to represent differences between classes and the common within class of generic objects is of great importance.So the 3D SIFT(scale invariant feature transform) descriptors of point clouds model based on the 2D SIFT is proposed.Then we propose a new algorithm based on multiple feature fusion of 2D and 3D SIFT descriptors respectively drawn from 2D images and 3D point clouds.The BoW(bag of words) model is used to calculate feature vectors and describe the objects according to the multiple feature fusion.The supervised support vector machine(SVM) classier is used to classify objects.Through several experiments,the advantage of this new algorithm is testified.

Key words: generic object recognition, point cloud, 2D SIFT, 3D SIFT, feature fusion, BoW, SVM

中图分类号: