In the process of image fusiontwo different-source video inputs often cause parallax between the two images due to the difference in optical system and spatial positionwhich in turn causes the problem of blurred edges and artifacts during image fusion.Thereforeimage registration is required before image fusion.Regarding the problem that the current heterogeneous image registration has poor real-time performance which makes it impossible to be applied in practical engineeringan image registration algorithm based on RANSAC and mutual information is proposedwhich divides the registration operations of different dimensions into the initial registration and real-time registrationin the meantime the POWELL optimization algorithm is introduced to speed up the searchwhich maintains a high real-time performance without reducing the registration accuracy.Finallythe proposed algorithm is transplanted to embedded hardware and accelerated based on Zynqand an embedded image registration system with high accuracy and real-time performance is realized.