As be approach the final weeks of the project, we’re getting initial results and need to begin diagnosing how we can improve them. The first step in this is plotting where our algorithm is calculating the camera to be at each frame.

As you can see, though the plot follows a correct generally circular path, it is at an angle, indicating we need to adjust the initial orientation. Initial testing shows that the initial orientation has a quite significant, and somewhat inconsistent effect on the final angle. This is likely because when the orientation the algorithm believes the camera to be at is different, but not the images, that has a more significant effect on how it calculates the change between the two perspectives for a given point, which results in a significantly different final position.

Using the orientation data from the metadata doesn’t seem to fix the issue, so we will likely need to experiment with the initial orientation parameter just with trial and error.