基于SIFT算法的图像目标匹配与定位

Matching and location of image object based on SIFT algorithm

针对仅应用SIFT算法无法准确定位的问题提出了一种将SIFT算法、马氏距离和仿射变换相结合的匹配定位方法。利用SIFT算法进行图像初次匹配,同时对初次匹配的特征点进行马氏距离计算,尽可能的剔除误配点,接着结合仿射模型计算出仿射变换的参数,最后根据目标模板的形心和仿射变换参数求出复杂环境中图像目标的形心位置坐标。实验结果表明了该方法在忽略几何畸变的情况下,对目标准确定位具有较强的适应性,能够实现在复杂环境中检测目标,并获取目标的形心坐标。

A matching and location approach is proposed and applied to solve accurate object orientation problem,which combines SIFT algorithm,Mahalanobis distance and affine transform.At first,SIFT algorithm is used to extract the first matched points,then Mahalanobis distance of the first match of the feature points is calculated and the mismatched points are eliminated,and then affine transform parameters are calculated with the affine model.Finally the object centroid coordinates in complex environment are obtained according to the object template centroid and affine transform parameters.Experimental results demonstrate that this method has strong adaptability and can detect the object centroid in complex environment in the condition of ignoring geometric distortion.