Image registration (the process of finding the mathematical relationship between pixels in one image to pixels in another) is an essential task in many computer vision applications; however, the problem is under-constrained and potentially unsolvable. The goal is to perform image registration on an FPGA in real-time, using as few resources as possible, and producing accurate frame-to-frame registration transformations between images. As embedded projects are almost never hardware independent, our goal is to clearly separate the vendor specific modules as to allow for relatively straight-forward porting to other platforms. Due to license restrictions, anything created with a vendor's wizard will not be able to be included in this project, and instructions/non-encumbered project files will be provided to allow legal creation of these files. It is not currently a goal of the project to be able to simulate everything intended for synthesis; however, this may change if there is a need for it. There will also be a high-level language version of the code available in C/C++ or Matlab (depending on the direction the project takes) that will produce results that will be compared to the hardware version for verification of each component.

New repo https://github.com/bwhite/fpga-image-registration