Auto White Balance Experiment

Theory

Here is an excellent article on what white-balance is and how auto white-balance (AWB) works, by robotics cameras manufacturer Point Grey. https://www.ptgrey.com/tan/11109

Basically, AWB works by forcing the "average colour" across the whole image to be gray (ie on the line between white and black). This tends to work well for photography because it cancels out tinted light sources like tinted bulbs or a setting sun by shifting all the colours to "neutral". This is not good for Vision in FRC because if we, for example, point the camera at a pile of cubes, we do not want the camera to "fix" the fact that the image is now full of yellow.

Experiment

The experiment was to determine how auto white balance affects Trackerbox2, and what cameras are best at tracking with auto white balance on.

There are 2 types of camera's tested:

• Microsoft LifeCam HD-3000
• Official Pi Ribbon Camera Model 2.0

With the Pi Ribbon Camera, Trackerbox2 was able to track the Power Cube until it took up about 60% of the screen, then it was unable to reliably track the cube. It also was running at ~5.5 fps and a latency of 1.5 Seconds. It is about the size of a nickel.

With the Microsoft LifeCam HD-3000, It was able to track the Power Cube until it took up the entire screen. It was able to track the entire time. It was running at ~4.8 fps and a latency of 1.5 Seconds. It is the size of a standard usb camera.

Questions:

• Prefer Distance of tracking vs. Size/type of connection of Camera
• How close does fab need to track up to