Image Matching Algorithm Based on Grid Acceleration and Sequential Selection Strategy
-
摘要: 特征点匹配可以在输入的图像之间生成一个对应关系,它是视觉里程计中一个基础且重要的模块,并且在不同的计算机视觉领域中有着广泛的应用。随机抽样一致算法(Random Sample Consensus, RANSAC)是一种应用较广的图像匹配算法,但存在召回率较低且耗时较长的问题。本文基于网格运动统计方法与顺序选取策略,提出了RANSAC改进算法。首先,对初始特征匹配进行质量排序,并在此基础上将输入图像划分为一定数量的网格,根据运动平滑性理论进行网格内匹配统计;然后,选取评分高的网格分别进行局部单应性矩阵估算;最终,将局部单应性矩阵进行聚合,进一步消除噪声影响,得到最优模型。此外,求取单应性矩阵时采用了顺序选取策略,进一步提升了算法的效率。仿真结果表明,本文方法具有较明显的优越性。Abstract: The feature point matching can generate a corresponding relationship between the input images. It is a basic and important module in visual odometry and has a wide application in different computer vision fields. Random sample consensus (RANSAC) is a widely used image matching algorithm, but it has the disadvantages of low recall rate and long time-consuming. By considering the grid motion statistics method and the sequence selection strategy, this paper proposes an improved RANSAC algorithm. Firstly, the quality of the initial feature matching is sorted, based which the input image is divided into a certain number of grids and the matching statistics in the grid is performed according to the motion smoothness theory. Then the grids with higher scores are selected to estimate the local homography matrix, respectively. Moreover, the local homography matrices are aggregated to further eliminate the influence of noise and obtain the optimal model. In addition, the sequential selection strategy is used to obtain the homography matrix, which further improves the efficiency of the proposed algorithm. Finally, the simulation results show that the proposed image matching algorithm based on grid acceleration and sequential selection strategy has better performance.
-
Key words:
- feature point matching /
- RANSAC /
- grid statistics /
- sequential selection strategy
-
表 1 SIFT特征点仿真实验结果
Table 1. Simulation experiment results of SIFT
Image group Algorithm Average precision Average recall Luminance variation RANSAC 0.813 0.782 RANSAC+GMS 0.863 0.824 Proposed algorithm 0.899 0.836 Scaling and rotation RANSAC 0.792 0.712 RANSAC+GMS 0.870 0.768 Proposed algorithm 0.902 0.818 Blur variation RANSAC 0.804 0.785 RANSAC+GMS 0.822 0.803 Proposed algorithm 0.869 0.846 表 2 算法耗时对比
Table 2. Comparison of algorithm time consumption
Number of
feature pointsTime consuming/ms RANSAC GMS+RANSAC Proposed algorithm 1000 80.70 53.34 58.78 2000 115.80 70.80 78.23 5000 240.10 123.16 135.89 10000 553.50 296.90 304.50 -
[1] 徐岩, 安卫凤. 基于改进随机抽样一致算法的视觉SLAM[J]. 天津大学学报(自然科学与工程技术版), 2020, 53(10): 1069-1076. [2] ZHONG M C, LU Y, QIAN J, et al. SLAM family receptors control pro-survival effectors in germinal center B cells to promote humoral immunity[J] Journal of Experimental Medicine, 2021, 218(3).ZHONG M C, LU Y, QIAN J, et al. SLAM family receptors control pro-survival effectors in germinal center B cells to promote humoral immunity[J]. Journal of Experimental Medicine, 2021, 218(3): e20200756. [3] 李晨玥, 张雪芹, 曹涛. 一种基于光度信息和ORB特征的建图SLAM[J]. 华东理工大学学报(自然科学版), 2021, 47(3): 331-339. [4] 王进祺, 高然, 董瑞虎, 等. 一种基于双目相机的SLAM系统的设计[J]. 数码世界, 2020(8): 65-66. [5] ZHAN H, WEERASEKERA C S, BIAN J W, et al. Visual odometry revisited: what should be learnt[C]//IEEE International Conference on Robotics and Automation (ICRA). USA: IEEE, 2020. 4203-4210. [6] RUBLEE E, RABAUD V, KONOLIGE K, et al. ORB: An efficient alternative to SIFT or SURF[C]//IEEE International Conference on Computer Vision. Barcelona, Spain: IEEE, 2011: 2564-2571. [7] LOWE D G, LOWE D G. Distinctive image features from scale-invariant keypoints[J]. International Journal of Computer Vision, 2004, 60(2): 91-110. doi: 10.1023/B:VISI.0000029664.99615.94 [8] BAY H, TUYTELAARS T, VAN G L. Surf: Speeded up robust features[C]//European Conference on Computer Vision. Heidelberg: Springer, 2006: 404-417. [9] FISCHLER M A, BOLLES R C. Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography[J]. Communications of the ACM, 1981, 24(6): 381-395. doi: 10.1145/358669.358692 [10] BARATH D, MATAS J. Graph-Cut RANSAC[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. USA: IEEE, 2018: 6733-6741. [11] LIU Y, GU Y, LI J, et al. Robust stereo visual odometry using improved RANSAC-based methods for mobile robot localization[J]. Sensors, 2017, 17(10): 2339-2357. doi: 10.3390/s17102339 [12] BRACHMANN E, ROTHER C. Neural-Guided RANSAC: Learning where to sample model hypotheses[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). Seoul, Korea: IEEE, 2019: 4321-4330. [13] BIAN J W, LIN W Y, LIU Y. et al. GMS: Grid-based motion statistics for fast, ultra-robust feature correspondence[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition. USA: IEEE, 2017: 1580-1593. [14] 杨晶东, 单体展. 基于网格的运动统计特征配准算法在医疗服务机器人中的应用[J]. 第二军医大学学报, 2018, 39(8): 892-896. [15] 孙凤梅, 胡占义. 平面单应矩阵对摄像机内参数约束的一些性质[J]. 计算机辅助设计与图形学学报, 2007, 19(5): 647-650. [16] 代文征, 杨勇. 基于改进高斯—拉普拉斯算子的噪声图像边缘检测方法[J]. 计算机应用研究, 2019, 36(8): 2544-2547, 2555. [17] 刘衍青. 双目视觉惯性融合的同时定位与语义建图研究[D]. 北京: 中国科学院大学, 2019. -