Buckinghamshire CC ANPR cameras

Roland Perry lists at internetpolicyagency.com
Tue Feb 7 14:02:09 GMT 2012

In article <20120207101536.54587f4c at peterson.fenrir.org.uk>, Brian 
Morrison <bdm at fenrir.org.uk> writes
>> >Since reg plates are not in one fixed position on vehicles it would need
>> >a pretty complex algorithm to find the plate
>> You have to do that to perform ANPR at all.
>> >and then look only at one part of it,
>> Wouldn't you "look at" all of it, then discard all but the middle?
>Well this is where the controversy over police requests to TM for their
>data came from, TM's description (that I can't now find) said that they
>simply tokenised the centre of the plate and then discarded the tokens
>after using them for flow analysis. I suppose that most people took
>that to mean that they didn't keep the raw plate images, but it would
>appear that actually they did.

If the camera is discarding all but the centre section of the plate, it 
would be consistent with both their traffic measuring objective but also 
being able to disclose to law enforcement a "trail" of particular centre 
sections[1]. But this begs the question of how the "centre section" is 
defined, because if it's not very precise (either in inches or in number 
of characters) then it wouldn't even work for the traffic measurement.

Given that some [vanity] numberplates are as short as two letters, the 
concept of "centre" might be a little strained.

[1] Using my earlier example, if you could show that a "centre section" 
of '?EN 1?' was consistently present on a route, you might regard that 
as sufficiently good signint about that particular car that day. There'd 
be collisions with plates like 'AEN 19' (rather than the target with P 
at the beginning and 5 on the end), but perhaps not catastrophically so.
Roland Perry

More information about the ukcrypto mailing list