2 EASY
WAYS TO
SUBSCRIBE
Visit
shop.kelsey.co.uk
/AP 1121
Call us on
01959 543 747
and quote AP1121
and d*Offer avaiigitallable teditionso new U, availableK based untsubscriil Novembebers, includr 30 2021 ing. Yprintou
will price. pay £Disco44.99 unts are ceveryalculat26 issues ed o– savn theing full 50% cover pofftrice ahe usual snd digitalhop
visit editioour wn downebsitload e at shofee.p.kelsFor oveyerseas.co.uk.For creddigiit/debtal it carsubscrid offptionsers, ,
at yourvisit shop.kelocal netlsey.cwork o.uk/arate. mateOrder urphotolinesgrapher.open8.Calls30am-charg5.30pmed ,
terms.Monday-KelsFridayey Pu. blishinFullT&Cg tas cankes your be foprivund atacy sershop.kiouslyelseyand .co.uk/will
and toonly use yproviour pde the ersonal productinfos anrmatiod servn to adices ministeyou haver yourrequeaccount sted
from upreferens. Wces yoe wiu will only cll indicatontact ye when oou aborderinut ourg ansped you cacial offen updrs via thate e
these us on 0at any 1959 time 543 by em 747. **Yailing oucanus at cunss@kelseubscribe y.co.uk ofrom the wer by calling ekly
newsletter at any time.
Save
50%
SPECIAL
CHRISTMAS
OFFER!
Get 26
issues
for only
£ 44. 99
Tech Talk
Professor Newman on...
Bob Newman is currently Professor of Computer Science at the University of Wolverhampton. He has been working with the design and development of
high-technology equipment for 35 years and two of his products have won innovation awards. Bob is also a camera nut and a keen amateur photographer
The eyes have it
O
ne of the interesting
features of the new
Canon EOS R3 is that
it restores a feature
last seen in the film EOS 3 of
1998 – that of eye-controlled
autofocus. Early reviews of the
camera indicate that the revived
technology works well, although it
requires calibration in order to
work for any individual user.
However, given the complexity of
navigating over 1,000 AF points,
the operational speed of the
system justifies its set-up time.
Since its announcement I have
seen a number of adverse
comments about the system,
possibly derived from Canon’s
marketing images. These show
beams of light being directed into
the eye, causing some people to
point out the danger of such an
arrangement, and conspiracy
theorists to suggest that Canon
is building a database of retina
patterns for identity tracking. To
find out more about how the
system works I took a look at the
patent covering it, Japanese
Patent Office 2021-076832. This
says little about how the system
tracks the eye, but a great deal
about how the eye is illuminated
and imaged. Directing focused
beams of infrared light into the
eye would indeed be a risk, but
thankfully, the system does not
do it. Instead it uses 12 LEDs to
provide diffuse illumination to the
exterior of the eyeball. This
implies that the conspiracy is not
happening; the system does not
image the retina. Also, it’s very
unlikely it could produce the kind
of database that is feared.
On to the part of the function
that the patent doesn’t cover,
how the eye position sensor
works. This kind of sensor uses
essentially the same technology
as an optical computer mouse.
The theory of operation is quite
simple: the sensor is a small
camera, in this case viewing the
eye via a beam splitter in the
viewfinder’s optical path. It takes
pictures of the object over which
movement is to be detected. It
then performs what is known as
a ‘correlation’ operation over the
images from successive scans,
moving the relative position of the
two images pixel by pixel until the
correlation between pixels is
maximised. The pixel offset then
gives the distance and direction
that the images have moved
relative to each other.
When I first encountered this
technology, which was developed
by Hewlett-Packard in 1998, I was
surprised by how basic the image
sensor was, with some using just
64 pixels. From the Canon patent
it appears that the R3’s sensor
has more like 1,000 pixels, which
might be one of the differences
between it and its 23-year-old
predecessor. Moreover, the pixels
are simple binary affairs, allowing
the correlation function to be
performed very simply. This can
be seen in the illustration below,
where an image of an eye has
been reduced to just 256 binary
pixels. The first two images are
the same eye with different lines
of sight, which means that the
pupil has moved. The remainder
are the results of a simple
difference operation between the
two images, with the alignment
between the two images moved
one pixel in each direction. The
one with the least difference
(i.e. fewest white pixels) has the
best correlation, and shows
where the pupil of the eye has
moved between the two.
A look at eye-controlled autofocus technology and
how it actually works
(a) and (b) are 256 pixel, 2 -level
images of a human eye. (c) is the
difference between the two
images. In the remainder the
image (b) has been displaced by
one pixel down (d), down and to
C D E F the left (e), and to the left (f).
A B
