All components in our high level design are functional and are performing as expected. An unexpected encounter was the ghosting effect which was never mentioned in any of the references we examined during our literate review. It is an unfortunate consequence of using an intensity-based stereo matching approach. We can only conclude that in previous projects, there is some kind of edge detection or feature extraction filter of some sort to distinguish between different objects, and compute a depth level for each object from the SAD approach. After trying out the basic SAD algorithm, it is very difficult to output a clear depth image without any major artifacts or noise. If we were to restart the project, there would be some kind of feature detection pre-processing filter used followed by a lightweight SAD module. We would also like to use an FPGA with greater logic capacity so that we do not feel limited by the window size or disparity ranges we are allowed to use.

Due to logic limitations and camera quality, we went through many iterations where we sacrificed image resolution, depth image clarity, and disparity range. Nevertheless, we managed to output a color-based as well as grayscale depth map image in real time which we felt was a great accomplishment.

Intellectual Property Considerations

All the code is generated without the use of any previously developed code, except for parts of the sample code for the camera modules provided by Altera and Terasic. We also used Skyler Schneider's VGA controller module, which was made available on the Cornell ECE 5760 web site. No reverse engineering of any kind was performed in this project. As mentioned before, there is little opportunity for patents or publications of any sort because depth mapping and stereo matching algorithms have existed for a long time. There are no legal considerations for our depth camera, as this technology is already sold and used by the general public.