To put it shortly, blood absorbs different frequencies of near infrared light depending on how much oxygen is being carried by its hemoglobin. Human bones are nearly transparent at these frequencies, so a head mounted device can measure the reflectivity of the outer layers of the brain right through the skull. The system can determine neural activity levels indirectly based off oxygen concentration, sort of like an fMRI.
The system is designed to shine one of two different infrared LEDs into the skull and measure the intensity of the reflected light. The two frequencies (730nm & 850nm) will be absorbed by hemoglobin with and without oxygen respectively. The reflected light will be captured by photodiodes, amplified, filtered and digitized with the ADC chip. The Atmega microcontroller will report results back to a host computer via USB. With 4 pairs of LEDs and 10 photodiodes clustered around them the system effectively has 16 measurement channels with 16 bits of resolution.
My project is released under GPLv3 and the LUFA library is provided under the LUFA license; both are available with my sourcecode in the github repository.