How did you get the idea?
While working on our robot Doki and our Adswitch
software (see IE website), we started to wonder about detecting patterns less
ambiguous than human faces. Using the same sophistication we had applied to
Adswitch, where we developed a library of complex algorithms and
trained artificial neural networks with hundreds of thousands of examples
of faces we wondered how much faster and more precise our detection could
be if the pattern was stable, exactly defined and with ideal contrast.
This lead to the design of our patterned holos sphere, for which we
were eventually able to achieve a 3D positional and orientational
detection accuracy many orders of magnitude better
than for human faces (angular: 1/15 of a degree, positional: 0.2 mm)
and also at 4 times the speed (60 frames per second) and
only using a fraction of the computational load (<12% of processor).
How long have you been working on this?
Our very first, very crude proof-of-concept prototype
for HOLOS vIWB with a huge sphere, took only a few months to build
and program and started to show potential already in November 2006 when
we applied for patents. At the time it was largely unclear however
whether all the large challenges regarding sphere size, detection speed,
detection range and accuracy, computational load, robustness
against background light, camera cost, battery duration etc.
could ever be solved to turn the idea into a viable product.
In the years since we have worked hard and managed to overcome all
the major obstacles step by step pushing up the performance while
at the same time reducing computational load and BOM costs to a
degree we had not thought possible initially and recently even
approaching military grade precision.