Figure 4: Given a detected point 𝑝 1 in the first image reference system, a point distribution is generated to obtain a set of multi-scale points 𝜆 𝑖 𝑝 1 . By using the predicted motion transformation 𝑀 𝑗 , they can be transformed to 𝑞 𝑖 in the second image reference system through 𝑀 𝑗 N( ̂ 𝜃 , 𝜙 , 𝜎 𝜃 , 𝜎 𝜃 ). Finally 𝑞 𝑖 are projected into the image plane to determine a restricted area where correspondences have to be found.