One of my forever projects is a full-size projected capacitive keyboard on glass or transparent thermoplastic sandwich with indium tin oxide electrodes.
(I.e., a fully transparent keyboard. A trinket, just like some people who love to wear a Rolex watch.)
Initially, I was hoping to use
NXP MPR121 chips, which are essentially 12-channel capacitance-to-digital converters, with input range from 10pF to 2000pF with a 10-bit ADC. Four chips can be connected to the same I²C bus (address 5A,5B,5C, or 5D, depending on where the ADDR selection pin is connected to).
I do not believe that a simple capacitance limit works well for touch detection. I was hoping to use the proximity data –– with 1 cm² electrodes, MPR121 can easily detect a finger at about 10mm distance –– and velocity data prior to the touch/limit event to detect a keypress. (I personally tend to have my fingers rest on the keyboard, so I'd want a "keypress" to really be a bounce away and back to the surface, tracked all the time, as opposed to just a "new" finger hitting the surface.)
Unfortunately, MPR121 isn't fast enough when it is precise enough, or precise enough when it is fast enough, for this: too much noise. It might be, with good electrode design, but me no EE.
So, what I really want now, is something like Apple Magic Trackpad, a projected capacitive proximity sensor, but in keyboard size and aspect ratio, essentially producing a high-resolution grid of capacitance values. At 100+ Hz, the velocity data would suffice for my esoteric purposes.
Anyway, if you end up doing a capacitive keyboard, with individual sensors for each key, I recommend you look at the hexagonal electrode patterns: it fits the QWERTY keyboard layouts very well, if you orient the hexagons with peaks at top and bottom, so left and right sides are vertical.