| 基于白化变换及曲率特征的3维物体识别及姿态计算 |
| 郑军, 魏海永 |
| 清华大学 机械工程系, 先进成形制造教育部重点实验室, 北京 100084 |
| Three-dimensional object recognition and posture calculations based on the whitening transformation and curvature characteristics |
| ZHENG Jun, WEI Haiyong |
| Key Laboratory for Advanced Materials Processing Technology of Ministry of Education, Department of Mechanical Engineering, Tsinghua University, Beijing 100084, China |
摘要:
| |||
| 摘要为解决3维物体识别及姿态计算问题，提出了一种基于白化变换和改进U弦长曲率特征的图像识别及姿态计算方法。该方法首先提取物体的2维形状特征，然后使用白化变换对模板物体图像轮廓和目标物体图像轮廓进行处理，使处理后的轮廓点集仅存在旋转关系；根据改进后的U弦长曲率方法，求取两轮廓的曲率，并进行匹配。实验结果表明：该方法具备较好的仿射不变性，其识别速度达到58 ms/帧（CPU：2.3 GHz；内存：4 GB），识别率在无遮挡情况下达到了100%，姿态检测精度达到了1.5°。 | |||
| 关键词 ：物体识别,仿射不变,白化变换,曲率,姿态计算 | |||
| Abstract：The whitening transformation and a U chord curvature are used to improve three-dimensional object recognition and posture calculation. The algorithm first extracts the shape characteristics of the object and then matches the contours of the target image with templates using the whitening transformation so that there is only a rotational relationship between the contour point sets. Then, the U chord curvature is improved to match the contours. Tests show that this method is affine invariant with a fast recognition speed which can reach 58 ms/frame (CPU: 2.3 GHz, RAM: 4 GB), a high recognition rate of 100% without shelter and a high detection accuracy of the posture calculation of 1.5°. | |||
| Key words：object recognitionaffine invariantwhitening transformationcurvatureposture calculation | |||
| 收稿日期: 2016-03-24 出版日期: 2016-10-25 | |||
| |||
| 引用本文: |
| 郑军, 魏海永. 基于白化变换及曲率特征的3维物体识别及姿态计算[J]. 清华大学学报（自然科学版）, 2016, 56(10): 1025-1030. ZHENG Jun, WEI Haiyong. Three-dimensional object recognition and posture calculations based on the whitening transformation and curvature characteristics. Journal of Tsinghua University(Science and Technology), 2016, 56(10): 1025-1030. |
| 链接本文: |
| http://jst.tsinghuajournals.com/CN/10.16511/j.cnki.qhdxxb.2016.22.033或 http://jst.tsinghuajournals.com/CN/Y2016/V56/I10/1025 |
图表:
 |
| 图１ U弦长曲率 |
 |
| 图２ 10幅模板图像 |
 |
| 图３ 缩放变换图像样例 |
 |
| 图４ 旋转变换图像样例 |
 |
| 图５ 剪切变换图像样例 |
 |
| 表１ 各方法的匹配实验结果 |
 |
| 图６ 各方法的缩放变换精度-召回率曲线 |
 |
| 图７ 各方法的旋转变换精度-召回率曲线 |
 |
| 图８ 各方法的剪切变换精度-召回率曲线 |
 |
| 图９ 各方法汇总数据的精度-召回率曲线 |
 |
| 图１０ 模板图像 |
 |
| 表２ 位姿检测精度 |
 |
| 图１１ 待识别图像 |
 |
| 图１２ 识别视频截图 |
参考文献:
| [1] 丁险峰, 吴洪, 张宏江, 等. 形状匹配综述[J]. 自动化学报, 2001(5)：678-694.DING Xianfeng, WU Hong, ZHANG Hongjiang, et al. Review on shape matching [J]. Acta Automatica Sinica, 2001(5)：678-694. (in Chinese) [2] Zhang D, Lu G. Shape-based image retrieval using generic Fourier descriptor [J]. Signal Processing：Image Communication, 2002, 17(10)：825-848. [3] 管云峰, 胡勇, 李介谷. 基于Fourier-Mellin变换的对称相匹配滤波算法[J]. 红外与毫米波学报, 1999, 18(6)：465-471.GUAN Yunfeng, HU Yong, LI Jiegu. Symmetric phase-matched filting algorithms based on Fourier-Mellin transform [J]. Journal of Infrared & Millimeter Waves, 1999, 18(6)：465-471. (in Chinese) [4] Tabbone S, Terrades O R, Barrat S. Histogram of Radon transform: A useful descriptor for shape retrieval [C]//19th International Conference on Pattern Recognition, 2008. Tampa, FL, USA：IEEE, 2008：1-4. [5] 刘云鹏, 李广伟, 史泽林. 基于Grassmann流形的仿射不变形状识别[J]. 自动化学报, 2012, 38(2)：248-258.LIU Yunpeng, LI Guangwei, SHI Zelin. Affine-invariant shape recognition using Grassmann manifold [J]. Acta Automatica Sinica, 2012, 38(2)：248-258. (in Chinese) [6] 杜海静, 肖阳辉, 朱丹, 等. 基于改进多尺度乘积LoG算子的仿射不变形状匹配算法[J]. 计算机应用, 2014(3)：841-845.DU Haijing, XIAO Yanghui, ZHU Dan, et al. Affine-invariant shape matching algorithm based on modified multi-scale productLaplacian of Gaussian operator [J]. Journal of Computer Applications, 2014(3)：841-845. (in Chinese) [7] 毛建旭, 刘敏. 基于ICA的仿射不变Zernike矩的交通标志识别[J]. 电子测量与仪器学报, 2013(7)：617-623.MAO Jianxu, LIU Min. Traffic sign recognition using ICA-based affine invariant Zernike moment [J]. Journal of Electronic Measurement and Instrument, 2013(7)：617-623. (in Chinese) [8] Arjun P, Mirnalinee T T, Sindhuja S, et al. Affine invariant shape descriptor using object area normalization [M]//Power Electronics and Renewable Energy Systems. Kolkata, India：Springer, 2015. [9] Arjun P, Mirnalinee T T, Tamilarasan M. Compact centroid distance shape descriptor based on object area normalization [C]//International Conference on Advanced Communication Control and Computing Technologies. Ramanathapuram, India：IEEE, 2014：1650-1655. [10] SHI Siqi, SHI Guangming, CHEN Xuyang, et al. Fast planar shape recognition using cascaded affine invariant function [J]. Infrared and Laser Engineering, 2012, 41(9)：2534-2542. [11] Sprinzak J, Werman M. Affine point matching [J]. Pattern Recognition Letters, 1994, 15(4)：337-339. [12] 吕玉增. 形状仿射不变特征提取与识别研究[D]. 北京：国防科学技术大学, 2008.LÜ Yuzeng. Affine Invariant Feature Extraction and Recognition of Shapes [D]. Beijing：National University of Defense Technology, 2008. (in Chinese) [13] 郭娟娟, 钟宝江. U弦长曲率：一种离散曲率计算方法[J]. 模式识别与人工智能, 2014, 8(8)：683-691.GUO Juanjuan, ZHONG Baojiang. U-chord curvature：A computational method of discrete curvature [J]. Pattem Recognition and Aitificial Intelligence, 2014, 8(8)：683-691. (in Chinese) [14] Heikkilä J. Pattern matching with affine moment descriptors [J]. Pattern Recognition, 2004, 37(9)：1825-1834. [15] 马金福, 薛弘晔. 基于Fourier-Mellin变换的图像配准算法及性能研究[J]. 计算机与数字工程, 2008(11)：134-136.MA Jinfu, XUE Hongye. Image registration algorithm and efficiency research based on Fourier-Mellin [J]. Computer and Digital Engineering, 2008(11)：134-136. (in Chinese) |
相关文章:
|
