[chair-law]

The algorithm dosent need to be too complex. You basically look at each still frame as grayscale, generate a hashtable of brightness values mapping to monospaced characters, and convert each grayscale pixel to a character.

[jefftk]

You can do better than that by taking advantage of character shapes, though. Minimal example:

    ()

and:

    )(

have the same brightness values, but we see them pretty differently.

Ideally you would use edge detection, maybe some object detection, and then be willing to compromise a little on brightness in order to make important edges nice. I don't know if there's software out there that does this, though.

[Caprinicus]

A better example would be ( vs ), checking neighboring pixels to weight that pixel correctly.

[chair-law]

How complex is an edge detection algorithm?

[sesqu]

Depends on how complex you want. Something like a Prewitt filter is super simple.