Here’s a really boring video of my airplane flying a circle hold for 26 minutes, but a lot went into making it this boring! 
A quick explanation: the camera is a standalone runcam hd 2 mounted above the nose recording to sd card … I just pressed record and flew. The HUD was added after the flight using python/opencv.
The position and attitude of the aircraft is logged in flight using a 15-state EKF developed by the University of Minnesota Aerospace Eng. department. The HUD is drawn (after the flight) using the real time onboard attitude/location estimate. The IMU is a ye-olde APM2 (mpu-6000) with a ublox7p gps. So all of this is done with very low end commodity hardware.
The camera calibration and lens distortion was derived using the usual technique of taking several pictures of a checkboard pattern and running through the opencv calibration solver.
I use some computer vision techniques to correlate the frame time of the video to the flight log, but when drawing the HUD, I’m just using the onboard attitude estimate. (I"m not using computer vision to find the horizon and do the alignment that way.)
The flight starts with the pitch estimate a bit too high, but after a minute or two it settles out pretty well. The sun is low in the sky and it’s hazy morning, so there is a lot of bleed though of light in the half of the sky centered around the sun and that affects the apparent horizon.
The sun/moon positions are computed using pyephem based on the flight log gps and time and updated dynamically through the flight. These can make really good absolute references for validating the alignment of the hud.
Hopefully enjoy being bored.