[Canberrauav] next steps for CanberraUAV
andrew at tridgell.net
Fri Oct 3 12:33:56 EST 2014
At our meeting on Tuesday evening we started a discussion on what is
next for CanberraUAV. There are lots of things we could work on, and
ideas from both inside and outside the team are very welcome.
>From my point of view I'd like to see us take a two pronged approach:
1) develop our own capabilities as a UAV S&R organisation
2) develop our software and documentation to support other groups
around the world to use the technology we have developed to allow
them to be more effective S&R organisations.
The two are complimentary of course, as developing our own capabilities
will tend to improve our software which will help others, but I think we
should have a real focus on making our technology more accessible to
others. We've put together some pretty useful bits of software as part
of our OBC development, but right now the software is largely
undocumented and quite tricky for other groups to use.
There are really three ways in which other groups can benefit from our
1) the work we do on the core ardupilot code benefits lots of people
immediately. We should continue to test and improve the code, with a
focus on features that help for S&R (the terrain following code is a
good example of that).
2) using our image recognition code as a post-flight analysis tool, via
our "geosearch" tool. This is the easiest way for other S&R groups
to use our code, as they just need an aircraft with any camera that
can store images on-board (eg. a point and shoot camera).
3) using our full in-flight recognition system, with real-time image
analysis, comms to the GCS etc. That is what we use ourselves, and
provides a fantastic S&R capability, but is much more complex to
setup and use, which means it needs even more documentation!
Apart from documenting our existing code for these 3 use cases, we
should look at how we want to improve the code.
For the OBC we just needed to find a highly distinctive human-sized
object from 100m altitude. When the OBC was originally conceived this
was an enormous task, but now it should really be considered quite an
easy task. We need to push the capabilities of the system a lot further
if we want to get something truly useful for S&R. In a real S&R
situation you can't rely on lost people wearing bright clothing and
lying out in the open. Instead the search involves looking for tiny
clues in the landscape that may help to locate the lost individual.
I think the new gold standard we should be trying to achieve is to be
able to find something around the size of a shoe over an area of about
10 square kilometres in one hour of searching. Right now our system
would have no chance of doing that, but I think that is the level we
should be trying to achieve if we want the system to save lives.
The justification for "shoe sized object" is that when people are lost
they may leave behind clues in the landscape. They may drop a water
bottle, a backpack, a hat or a jacket. We need to not only find those
objects, but be able to display a high enough resolution image of them
on the GCS that the search coordinator can decide if it is worth sending
out a ground team to investigate.
To achieve that we are going to need much better cameras and much
smarter software. From an altitude of 100m the pixel size of our PtGrey
Chameleon camera is about 10cm at the moment, which is far too large for
finding shoe-sized objects. We could find shoes if we flew much lower,
but then we wouldn't be able to cover large areas.
The good news about this sort of target capability is that it should be
reasonably easy to test at our existing flying field at CMAC. We can
scatter small objects over quite long distances and try to find them
while staying within the flying limits of our field.
Apart from higher resolution and smarter software we can also branch out
in other directions. For example, looking in the near infrared, and
using thermal cameras. NIR cameras should be quite easy to try and not
terribly expensive. Thermal cameras are much more expensive (and low
resolution), but offer huge advantages in real S&R situations.
I think that is enough to get the discussion going, please post your own
ideas! I'll write a separate post on some specific hardware I think we
More information about the Canberrauav