[1]贾一鑫,邓魏永,殷 强,等.基于改进 SURF 的增强现实图像匹配方法[J].计算机技术与发展,2024,34(01):59-64.[doi:10. 3969 / j. issn. 1673-629X. 2024. 01. 009]
　JIA Yi-xin,DENG Wei-yong,YIN Qiang,et al.Augmented Reality Image Matching Method Based on Improved SURF[J].,2024,34(01):59-64.[doi:10. 3969 / j. issn. 1673-629X. 2024. 01. 009]

点击复制

基于改进 SURF 的增强现实图像匹配方法()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

34

期数:

2024年01期

页码:

59-64

栏目:

媒体计算

出版日期:

2024-01-10

文章信息/Info

Title:

Augmented Reality Image Matching Method Based on Improved SURF

文章编号:

1673-629X(2024)01-0059-06

作者:

贾一鑫¹ ; 邓魏永² ; 殷 强³ ; 毋 涛¹

1. 西安工程大学 计算机科学学院,陕西 西安 710699;

2. 华宇铮蓥集团,福建 泉州 362801;

3. 中国纺织工业联合会,北京 100020

Author(s):

JIA Yi-xin1 ; DENG Wei-yong2 ; YIN Qiang3 ; WU Tao1

1. School of Computer Science,Xi’ an Polytechnic University,Xi’an 710699,China;

2. Huayu Zhengying Group,Quanzhou 362801,China;

3. China National Textile and Apparel Council,Beijing 100020,China

关键词:

增强现实; 图像匹配; SURF 算法; DAISY 描述符; 随机抽样一致; 三角不规则网络

Keywords:

augmented reality; image matching; SURF; DAISY descriptor; random sample consensus; triangular irregular network

分类号:

TP391. 41

DOI:

10. 3969 / j. issn. 1673-629X. 2024. 01. 009

摘要:

针对传统增强现实图像匹配算法鲁棒性不强且效率不高的问题,提出一种改进的 SURF 匹配算法。 首先,使用SURF 算法进行特征点检测,并通过 Haar 小波模板确定特征主方向,在得到特征主方向后构建特征描述符;由于传统SURF 算法采用高达 64 维的矩形描述符,导致算法的计算量非常大,并且鲁棒性不强。 因此,该文使用 DAISY 圆形描述符替代原始算法中的矩形描述符,DAISY 是三层同心圆结构,每层包含 8 个采样点,可以得到 25 个维度的描述符,这种结构使得算法的鲁棒性大大增强并且降低了计算复杂度;接着,使用特征描述符计算欧氏距离进行特征点匹配;最后,对得到的匹配点集使用随机抽样一致( RANSAC) 与三角不规则网络( TIN) 算法进行优化,剔除误匹配点。 实验结果表明,该算法虽然略微增加了时间复杂度,但鲁棒性变得更强,并且算法的效率和匹配精度也大大提高,平均精度达到了 95% 以上。

Abstract:

Aiming at the problems of poor robustness and efficiency of traditional augmented reality image matching algorithms, animproved SURF matching algorithm is proposed. Firstly,

the SURF algorithm is used to detect the feature points,then the Haar wavelettemplate is used to determine the main direction of the feature,and then the feature descriptor is constructed after the main direction of thefeature is obtained. Because the traditional SURF algorithm uses rectangular descriptors up to 64 dimensions,which is computationally

intensive and robust. Therefore,the DAISY circle descriptor is used instead of the rectangle descriptor in the original algorithm. DAISY isa three - layer concentric circle structure, each layer contains 8 sampling points, which can obtain 25 dimensional descriptors. Thisstructure greatly enhances the robustness of the algorithm and reduces the computational complexity. Then Euclidean distance is computedusing the feature descriptor for feature point matching. Finally,the resulting set of matching points is optimized using random samplingconsistency ( RANSAC) and triangular irregular network ( TIN) to eliminate mismatched points. The experimental results show that thetime complexity of the proposed algorithm is slightly increased,the robustness becomes stronger,and the efficiency and matching accuracyof the proposed algorithm are also greatly improved,with the average accuracy reaching more than 95% .

常用功能