Thoughts on long-range sensing
Guy
ideas on long-range sensing. I picked up a dollar store laser pointer
($1.50 actually) and played with it last week and I just realized that
a camera should be able to pick out it's spot quite easily. So my
thought is to use 2 laser pointers aimed in parallel with a camera and
by counting the pixel offset between them and between the camera
centre we can get a distance estimate for whatever is in the centre of
the gaze. Gaze is important here so we will probably want to create a
basic neck using servos so that we can move the gaze without needing
to turn the vehicle.
I've got 2 standard and one micro server on order from Tower Hobbies
and I'm about to order an AVRcam. One of the dollar stores in
Lougheed Mall has plenty of the laser pointers so all that remains is
to build, program, test, and integrate. If the ATMega8 on the AVRcam
can handle the servers along with communications with the base
processor then we should be set. Otherwise we may need an additional
processor.
Anyone think this a good idea and want to take it on? We've only got
about 6 weeks before the fair.
BTW, I'm also planning to order some ultrasonic sensors from
Robotshop.ca so if there's anything that you'd like to add to the
order let me know.
-- Guy
Jonathan Hanna
> I did some research on sensors last night and I've come up with some
> ideas on long-range sensing. I picked up a dollar store laser pointer
> ($1.50 actually) and played with it last week and I just realized that
> a camera should be able to pick out it's spot quite easily. So my
> thought is to use 2 laser pointers aimed in parallel with a camera and
> by counting the pixel offset between them and between the camera
> centre we can get a distance estimate for whatever is in the centre of
> the gaze. Gaze is important here so we will probably want to create a
> basic neck using servos so that we can move the gaze without needing
> to turn the vehicle.
Keeping the beams reasonably parallel should be possible for short distances, such
as 5 meters, beyond the range of the other sensors. It may need a fairly sturdy
mount. Laser pointers are rated as eye-safe, but I avoid them and am not
really happy that they are available at dollar stores. Another possibility is to
point them at the ground ahead, determining if there is is safe to go
at a reasonable speed. I am not sure if the spot is really that easy
to pick up. It may be, and should be easy enough to test.
> I've got 2 standard and one micro server on order from Tower Hobbies
> and I'm about to order an AVRcam. One of the dollar stores in
> Lougheed Mall has plenty of the laser pointers so all that remains is
> to build, program, test, and integrate. If the ATMega8 on the AVRcam
> can handle the servers along with communications with the base
> processor then we should be set. Otherwise we may need an additional
> processor.
>
> Anyone think this a good idea and want to take it on? We've only got
> about 6 weeks before the fair.
I could test the AVRcam with laser pointers and see if anything needs
to be modified in its blobification. I don't think I would be able
to also handle servo control, unless it is dead easy.
Peter Smith
We used a laser level from home depot to produce a laser line and then slowly scanned the line over objects to build a 3d representation using the parallax method similar to what guy was describing. Detection was with a $30 logitech web cam. Image processing was done in Matlab.
We were only scanning about 1 m away but it did a fairly good job of generating range data.
However, we found that the detection was difficult. We had to do the process in a darkened room and use a two frame technique. We would grab a frame with the laser off and then another with it on. By taking the difference between the two frame we could get a reliable detection of the reflected laser light.
Also, the image processing was a bit slow. If memory serves, it took a minute to do a scan along a couple of meters. Mind you we were using a 5 year old PC.
I think the idea has merit, however to be usable in broad daylight with a range of several meters, a more sensitive and specialized (and expensive) camera needs to be used. Perhaps using an optical filter would help. Also, specialized hardware would probably have to be used to speed up image processing.
Another idea is to modulate the laser with a signature signal and then use a digital filter to look for the signal in the received image, though this might need some hefty computing horsepower or some specialized circuitry.
just my 2 bits
-peter
-----Original Message-----
> Date: Fri Jun 13 12:51:45 PDT 2008
> From: "Jonathan Hanna" <jha...@pangolin-systems.com>
> Subject: [Starfox] Re: Thoughts on long-range sensing
> To: "Team Starfox" <team-s...@googlegroups.com>
Jonathan Hanna
If the AVRCam cannot easily pick up laser points then I think only simple things like optical filters are really worth pursuing.
A significant amount of processing power is probably better spent on techniques such as Stanford's
monocular depth perception (http://cs.stanford.edu/people/asaxena/rccar/).
Any ideas about simple modulations? The camera can give 30 frames a second, I think. Even standing still in the real world there will be a lot
of differences in frames. Going to space instead of time the lasers could be made to wiggle real fast to produce a pattern that
an image filter may pick up, but that sounds like a big engineering problem.