Analysis of multi-resolution image registration algorithms in real time applications

As machine vision develops and shows increasing power in solving problems, it is given a more important role in many applications, such as teleoperation. One of the proposed systems is to have wide field of view and high resolution at low cost, by using slaved high resolution insert on top of a low resolution wide field of view background. In such a system, there is a need to register images with rotational/scale/translational (RST) changes introduced by two different cameras. Then two pictures are fused or projected to give a wide field of view with high resolution insert in the area of interest (AOI). For real time implementation of such software, there are a number of factors that have an impact on the performance index such as the estimation accuracy, reliability, and the computational cost. Our analysis show that the relations between these factors and the estimation performance can be quantified and used as guidelines in the implementation. Both a correlation-based approach and a feature-based approach are investigated in this paper.