| |
Having an innovative idea such as Dream
Recorder is a first step. But bringing it to life requires a
lot of technology and data processing. The osX framework offers
the ideal technology more than any other platform. We brought
the know-how and inventiveness. Dream Recorder has been
designed and realised by professionals of the field of
scientific instrumentation.
OsX Embeded Technologies: Interfacing a computer with
humans requires a lot of technology. The Apple know-how in that
field is still unbeaten. OsX was the ideal environment for
developing such a project. For developers like us, Core Audio
and Core Image are outstanding toolkits taking care of most
data operations in an efficient manner. Just invent and forget
about optimization... Indeed, keeping the processing power low,
thus avoiding fan noise, while implementing complex real-time
image and audio processing is a challenge.
Night Vision Processing: Dream Recorder uses an image
processing algorithm that stems from professional astronomers,
using large CCD for deep star fields observation. A CCD sensor
yields intrisic perturbations called the dark current and
read-out noise. This makes the images looking grainy in dark
environments.
On successive iSight images, the noise pixels will be different
as they are random. On the other hand, real information will be
present each time. Dream Recorder analyses the correlation
between pixel information, image after image, and filters the
noise by calculating the statistical variance.
Moreover, the image motifs are spread above many pixel. Here
again, neighbor pixels will keep the same contrast, giving an
additional noise rejection criteria.
Dream Recorder typically takes 60 dark images and produces a
rendered 3D model used for motion detection. This methods works
only on static subjects, which is of course the case of a
sleeping person. This would not be efficient when filming
moving subjects.
Voice processing: Dream Recorder permanently analyses
each sounds of the night. It recognizes them and makes the
difference between a person talking and snoring.
When you tell about the dream you just had, the sound
processing will automatically record your voice and log it for
you, without even touching the keyboard. After some seconds of
silence, Dream Recorder will switch back to survey mode.
In the fisrt version of Dream Recorder, snoring was not
distinguished from voices. Our users had gigabytes of snoring
logged on their hard disk... Today's versions can just skip
snoring. The amount of snoring is logged for statistics, and
some snoring whistles can also be emmited by Dream Recorder...
even if this feature is more a kind of gadget, but appreciated
by some of our user's partner.
The voice processing is done throught the analysis of sound
spectrograms, using the Core Audio framework of osX. This
spectrogram is these of a person snoring. It shows some
repeting patterns, at the rithm of the sleeper's
breath.
|
|
|