摘要
针对基于特征匹配的目标识别算法复杂度高、难以实时处理的问题,提出基于快速鲁棒性特征(SURF)的快速特征匹配算法。通过应用双阈值顺序聚类算法对特征点进行聚类,并对每一个聚类建立k-d搜索树,采用优先搜索算法匹配模板与图像的特征点,提高了算法实时性。采用RANSAC鲁棒估计算法消除错误匹配点对,计算模板与图像平面之间的单应矩阵,进而实现对目标的准确识别定位。仿真实验证明了算法的有效性和实用性。
Considering that the target recognition algorithms based on feature matching has high computational complexity and it＇s hard to be processed in real time, a fast speeded up robust feature （SURF） matching method is proposed. First, the real time image SURF features are extracted, and the two-threshold sequential algorithm scheme is applied for feature point clustering. Then, one k-d search tree is established for each cluster for BBF priority search to match template and image features. Finally, the RANSAC robust estimation algorithm is applied to eliminate the error matching points and estimate the homography of template and the image plane. Experiment demonstrated the validity and practicality of the algorithm.
出处
《电光与控制》
北大核心
2013年第4期68-71,共4页
Electronics Optics & Control
关键词
目标识别
图像匹配
顺序聚类
优先搜索算法
单应矩阵
target recognition
image matching
sequential clustering
priority search
homography
