Searching \ for '[EE]: Update on determining speed' in subject line. ()
Help us get a faster server
FAQ page: www.piclist.com/techref/index.htm?key=update+determining
Search entire site for: 'Update on determining speed'.

Exact match. Not showing close matches.
'[EE]: Update on determining speed'
2001\02\15@004534 by

Hi all,

Thanks to everyone who responded to help me with my assignment to determine
car speed from the audio recorded by a stationary observer as the car passes.

What I did was to go out and make recordings of cars passing using my
camcorder. The camcorder records stereo audio as well as video. From the
video, I was able to determine the speed of the cars, so I now have what I
wanted, several samples of audio where I already know how fast the car was
going. Thanks to Olin for this great idea.

Well, I tried to get the doppler idea to work, but really the only sound I
get from most cars is the whoosh of the turbulent air around them as they
move (although, I did get some strange looks, standing by the side of a
road with a camcorder ;-) Since this is a very broadband signal, I haven't
been able to detect any usable doppler shift when comparing the spectrum
from before it passes to after it passes.

So, I moved on to a method using the amplitude (modeling the car as a point
source, assuming that the inverse square law holds, and trying to determine
how fast it would have to be going to produce the given change in amplitude
over time as it approaches or recedes) . At first this seemed to work well,
but it doesn't seem to work equally well for all the sound clips, so I am
still working on it.

The idea, proposed, I think, by Peter Peres, to use the stereo nature of
the recording to determine speed may also have some merit. I had dismissed
it earlier since I don't have information on the sensitivity pattern of the
microphones on my camcorder. However, I may be able to calibrate the system
using one a sound clip where I know the car's speed, and then test the
system on other sound clips.

The project is actually due on Tuesday, so I will give an update when it is
finished. If you have any further suggestions, please tell me.

Thanks,

Sean

--
http://www.piclist.com hint: To leave the PICList
piclist-unsubscribe-requestmitvma.mit.edu

> Well, I tried to get the doppler idea to work, but really the only
> sound I get from most cars is the whoosh of the turbulent air
> around them as they move
> Sean

Can you get someone to drive by and sound their horn ? Or
hold up a "Honk If You Like Sex" poster ?

--
http://www.piclist.com hint: To leave the PICList
piclist-unsubscribe-requestmitvma.mit.edu

You may be running into amplitude compression as the camcorder's
automatic gain control operates on the rising sound pressure. You need
to force a fixed gain or supply a reference tone (piezo buzzer?)
whose amplitude you can measure, as the AGC operates, to calibrate
out the gain change.

Also, look at the spectrum -shape- since you have no clear tone
to extract. The overall shape shouldn't change much, but will
shift in frequency. Also, I assume that you're using a logrithmic
scale to extract the low level information buried under the
wind noise.

You are probably also recording from too high a
location. The mufflers are UNDER the car so recording from curb
level should give you more tonal content, as well as buzz from
the tires.

I spent half an hour with a LIDAR equipped cop who was happy to
show off the accuracy and ease of use of their new toy. (LTI 20/20).
I was amazed at the pinpoint aiming accuracy, and how after a
few minutes practice I could guess a cars speed to within 2km/hr.

I also learned that license plates are their favourite target
(because of the retroreflectivity) and that the units are smart
enough to reject rapid range changes such as happen when the beam
pops from the bumper to the grill to the hood to the windscreen
as they pick off cars coming down a hill. It certianly made me
appreciate how hard it would be to argue 'flawed use' in court.

A friend of mine used to calibrate/repair radar guns for the RCMP
& city police, and would test them on a nearby freeway, much to
the consternation of the local truckers who would lock up their
brakes as he keyed the unit just as they were within range.
He too related how one can quickly learn what a vehicle 'sounds'
like at whatever speed it passed.

And the other way to do this is cheat. Since you have two channels
on your camcorder, use two microphones spaced a distance apart.
Time between amplitude peaks will get you the speed without using
doppler. (Or in captain Kirk terms, 'change the program' [Kobiashi
dilema]).

You could also look at correlating the audio between the two
channels to measure the wavefront location, and it's change with
time. Cross correlation is more quickly handled by doing an FFT,
then a complex multiply between channels. The amplitude peak occurs
at the delay corresponding to 1/Tau (if memory serves. It's been
20 years since I did this for nerve recordings).

Robert
"Sean H. Breheny" wrote:
{Quote hidden}

--
http://www.piclist.com hint: To leave the PICList
piclist-unsubscribe-requestmitvma.mit.edu

> The project is actually due on Tuesday, so I will give an update when it
is
> finished. If you have any further suggestions, please tell me.

Maybe a little late -
What about making an audio spatial comb filter so that the sound
successively reaches the microphone or is blocked as the car passes.

|
Car   *
| z   |     *
\/      z        *
|         z     *
x               z *
|                 x     Mic

|

The front noise edge of the car would cause a series of pulses which were
dependant on the rate at which the vehicle passed.

Also, you may well get good correlation (possibly very good) as the vehicle
successively presented itself in the same relationship at successive sets of
slot openings. I suspect I'm not making this very clear :-) !!!
eg

Part of car        Slot seen through

Timeslot       1    2    3    4    5    6    7    8    9    10    11

Front                    1    2    3    4    5    6    7                 <-
physical slots
f door                         1    2    3    4    5    6    7
r door                               1    2    3    4    5    6    7
rear                                         1    2    3    4    5    6    7

Initially only the front is seen through physical slot 1
After a period t the front appears in physical slot 2, the front door in
physical slot 1 etc
After 4 x t the car appears in slots 1/2/3/4
After another t it appears in slots  2/3/4/5
This continues up to timeslot7 when it appears in slots 4/5/6/7
After this the front moves out of picture and the car "fades away"

Looking for the time between correlation peaks gives the car speed.

This would allow a single microphone mixing all sounds.
Alternatively each "physical slot" could instead be a separate microphone
recorded separately.
Correlation between these should be even easier but the sound collection

Any good?

regards

Russell McMahon

--
http://www.piclist.com hint: To leave the PICList
piclist-unsubscribe-requestmitvma.mit.edu

>(although, I did get some strange looks, standing
>by the side of a road with a camcorder ;-)

It must have looked like some kids here in England who were fed up with cars
speeding through the village where they lived. They put on some hi-visibility
jackets and stood out on the road side pointing a small hand held hair dryer at
every car going past. Apparently produced some rapid deceleration tests in a few
cars for several days.

--
http://www.piclist.com hint: To leave the PICList
piclist-unsubscribe-requestmitvma.mit.edu

> Well, I tried to get the doppler idea to work, but really the only sound I
> get from most cars is the whoosh of the turbulent air around them as they
> move (although, I did get some strange looks, standing by the side of a
> road with a camcorder ;-) Since this is a very broadband signal, I haven't
> been able to detect any usable doppler shift when comparing the spectrum
> from before it passes to after it passes.

Did you use the whole spectrum or part of it?  I would guess that the
unwanted noise is of higher frequency than the sounds being emitted by the
car itself.  How about isolating a peak or two at the low end of the
spectrum, then tracking only those peaks as they shift in frequency?

*****************************************************************
Olin Lathrop, embedded systems consultant in Devens Massachusetts
(978) 772-3129, olinembedinc.com, http://www.embedinc.com

--
http://www.piclist.com hint: To leave the PICList
piclist-unsubscribe-requestmitvma.mit.edu

More... (looser matching)
- Last day of these posts
- In 2001 , 2002 only
- Today
- New search...