SKY REPLACEMENT & COLOR GRADING
Original :
Objective :
To Replace the Sky in the footage and color correct it to match with the original.
Technique used :
Match Moving (3D)
Luminance extraction (Alpha Matte)
Color Correction
Challenges :
a) The camera has a rotational motion.
The car is moving, so it should be excluded from the Tracker.
To replace the sky, it has to be perfectly tracked.
The error value of the track should be below .7.
b) There are lots of holes between the branches of the tree. Proper extraction of Alpha channel is a must.
c) The light from the source is from the right side of the footage. Appropriate image should be selected for the replacement.
The Process:
First, I created a ROTO shape for the moving vehicle, animated the shape to follow the vehicle's movement throughout the footage.
Then I input the data into 3D CAMERA TRACKER.
This ROTO shape will exclude the moving vehicle when the 3D CAMERA TRACKER is tracking the footage, which in return will give a proper/accurate camera solve.
The error value was too high, so I increased the "min track length".
I also decreased the "max error" and "max track error."
This will eliminate a lot of bad tracks which will result in better solve error.
Now that the camera is solved I created the scene.
I am using a "sphere" 3D model to project the pixels of the replacement sky.
I can use a 3D Card which has only x and y axis, but since the movement of the camera is rotational, I am using a "sphere".
Moving on to the Alpha Matte extraction
The solution here is to use a "Keyer" node which has the luminance slide.
These slides can be adjusted to extract the alpha matte,
Sky pixels = Black
Other pixels = White
I adjusted the sliders to create the "Alpha Matte"
I am faced with the next challenge which is the dark spots inside white.
The simple solution is to "Channel merge" a "roto" mask on it.
Next step is to "merge" the 3D Scene with the replacement sky image on sphere and the Alpha extracted footage.
Also added "grade" node to match the Balck and White points in both the footages.
Finally used a "color correction" node for matching the tone of both the footages
Finally, here is the Before and After
