Drift Issues Continued.

I watched a tutorial on digital tutors to try and work out the causes of the drift. I found out a few things about the track I did.

1. My track error was above 1. This would mean an inaccurate track. 
2. Because you can't select individual points in the point cloud in Nukex it makes placement quite difficult. This would explain why people like to select their own points rather than auto tracking. 
3. My detection thresh hold was too high, I was tracking 150 features inaccurately

To sort these issues I;

1. used the refine tool to refine tracks and reduce the error to below 1. 

2. Picked obvious points that referred to things i knew in the footage. For example, the few floor tracks. 

3. I upped the amount of points and increased their accuracy. I ended up with the same points but they were better.

This is the video changing these settings. 

As you can see the track is far better. But I know this is more down to it being quite easy to do. The ground plane in this video is still off but because this is a sphere it doesn't notice so badly. I am going to attempt a similar experiment in Boujou and see how the process compares.