摘要
提出了一种基于图像不变特征的目标匹配算法.算法首先采用了一种改进的SIFT图像特征点提取技术提取目标的SIFT特征向量;建立改进的Kd-Tree特征结构,使用BBF搜索策略完成特征的匹配,接着建立目标的姿态变换空间对匹配点进行HOUGH聚类,去除错误的匹配点,最后对匹配点按照最小二乘法拟合出目标的姿态参数,从而完成目标的定位.实验结果表明,在目标发生平移、旋转和缩放以及场景部分遮挡、视角变化等因素引起的图像变形时,算法均能够稳定地匹配出目标.
A target matching algorithm based on Invariant features is presented. First the algorithm extracts the the SIFT features using an improved SIFT points feature descriptor, and establishes the modifying Kd-Tree structure, and then complete the point match useing BBF. after that it establishes the posture space and uses the HOUGH cluster to remove the outer points, and then locates the object uses the LM algorithm. Experiments show that the algorithm can recognize the object across a substantial range of scale change, rotation, change in 3D viewpoint and so on.
出处
《华中师范大学学报（自然科学版）》
CAS
CSCD
北大核心
2010年第1期38-42,共5页
Journal of Central China Normal University：Natural Sciences
基金
广东省本科高等教育教学改革项目(BKJG200765)
