Correspondence
A new approach to 3D object’s feature points correspondence is presented by H.-Y. Zhang et al. (2003). It has 3 steps. First, we evaluate the pose by matching the input image with reference image
6
using the aspect graph. Next, polar-exponential grid sampling and polar-log coordinate map are used to approximate feature point location of the moving target. Then the correspondence can be obtain by getting the means. This approach solve the problem of correspondence on plane object moving.
H. Zhang et al. (2010) present an improved 3D object feature point correspondence algorithm. Its methods can be classified into 3 steps. First, an object model database that based on aspect graphics is builted by an ART neural network, and quick search object model database algorithm is then proposed. The second and third steps are same as H.-Y. Zhang et al. (2003) method. This method solves the problem object model database having so many images and hard for searching.