Object Matching

One of the simplest approaches to object identification is the alignment method. We represent an object by m points u₁,...,u_m. An object in a scene is rotated by a rotation R and translated by a translation t. We look for n corresponding image points p₁,...,p_n, such that:

p_i = P_i(R u_i + t), or
p_i = Q u_i

We can find Q given three sets of u-p points. We keep trying triplets of points until a good Q is found. A good Q maps the remaining points well.

This procedure is expensive for many possible objects.