Ultrasonic 'Vision'?

[mechie]

Has anybody out there played with ultrasonics for robot vision?

I'm playing with a lash-up, looking at measuring the time between sound pulse out and echo back (around 1.75mS/foot if you're interested)- nice and easy for things at right-angles to my transducers :!:
I intend to measure upto a few feet and get some sense of direction by sweeping the transducers (or a director).

There is some correllation between object size (and hardness and angle of approach) to echo strength as well- I'm considering utilising this info as possible 'texture analysis'.

If bats can do it then maybe we can?

 

[giaosucan]

you can visit website **broken link removed**
you can find some info about ultrasonic .
I also have some documents about that .If you need ,you can contact me
have fun

 

[jbeng]

Hey Mechie, I hope you can read the language that forum is written in... :shock:, I sure can't.
Is that forum available in English? Just curious...

JB

 

[bmcculla]

If you are really hard-core you could build a phased array ultrasound sensor. Using say 10 ultrasound transducers in a line you can vary the phase of the driver to each of them to sweep the ultrasound beam back and forth with no moving parts. The idea is the same as with the wave theory for optics. A lens phase delays the wavefront due to varying thickness which focuses light. You can fake this effect with your transducers by creating a different phase delay to each transducer. This lets you electronicaly sweep the beam back and forth. High end radar systems do this to get rid of the moving dish. It wouldn't be too hard to use a cheap FPGA to create a buch of phase settable ultrasound drivers.

Bats use an effect similar to this for their echo location. They measure the phase difference in the received sound between their two ears to locate objects.

 

[dknguyen]

I was considering such a project too. YOu might want to use wide-bandwidth sonar transceivers so that you can use chirps and other multi-frequency tricks to increase your resolution at close range and to get around how some objects supress certain frequency echos more than others...its expensive though. I was look at an mBAT transducer- (MHz-range widebandwidth!) but I guess its research level so it cost a ton ($1500 which is in the range of possiblity, I guess). But because it didn't work like regular transmitters and receivers it required some unique drive circuitry which I don't know enough to build myself and buying that from the company pushed it up to nearly $10k...I figured it would be cheaper for me to buy some cameras and try and work on a crude visual obstacle avoidance system by frame differencing stereo-images and time-differencing the frames...which is one of the monster Im trying to figure out now.

Have you read up on how bats do it? Its seriously cool stuff. They do better than the best military sonar. Search up a bunch of articles online and see some of the things they do and see if you can't use them. You probably aren't moving fast enough to have to compensate for the doppler effect liks bats do though hehe...but if you do, you will need a more costly wide-bandiwdth transmitter/receiver.

Beamforming which uses constructive and destructive interference (correct me if I'm wrong) to sweep and make narrower beams (mentioned by bmculla). It is something the navy uses that you could use if the sensors you go with dont cost too much. THe potential price you pay is that you can't use wide-bandwidth transmitters/receivers for multi-frequency operation- if you go that route would probably use a single transceiver and a pan-tilt unit.

Advantages of multi-frequency operation is that you can get around how some objects absorb some frequencies more than others. If you use a multi-freqneyc chirp instead of a constant-frequency tone, its also easier to pick out in noisier environments. Also, if you can adjust the frequency you can adjust the beamwidth- increase the frequency (when needed) to decrease beamwidth and increase resolution at the cost of reduced distance (but that kind of makes sense, if you are looking at something very closely, you can't see it from very far).

Phase difference is also something you should definately look into. It's kind of like how humans detect the time difference that it takes for a sound to reach one ear after the other- its sort of unconcious to us but it lets us figure out where a sound is coming from without the need for knowing how loud a sound should be. You can use it to augment the facy of a known beamwidth narrow down where an echo is coming from. And you would be able to control the beamwidth if you use a widebandwidth sensor and control the frequency...however, you have to try and deal with the problems where one large flat wall can give off multiple echos because it is so large. Maybe you can correlate the signal in the different receivers to phase-difference them by looking at similar echo-frequency characterstics AND similar dB power levels?

Another thing, this is practically a project all its own, but if you can build a ultrasonic echo sound library and figure a way to correlate real-life samples to it, you might be able to guesstimate what certain objects are or extrapolate the surface properties. You could then maybe use the phase-differencing to narrow down what direction it came from out of all the echos you receive. But it's high speed stuff...you need to be able to pick a specific type of echo quickly out of all the noise you get back fast enough so that you can measure the time difference between different receivers...not to mention it probably won't sound exactly the same to all receivers. That's the problem with surface texture- it requires "prior" experience and the only way to do that is with a library (or...AI!, but then you need other complex sensors to feed in tactile information)...as far as I know there are no general mathematical rules you can use (except maybe softer object makes softer echo, but that is also dependent on wind and distance, if you can work that out). I am not quite sure how a 40kHz echo off different materials will differ in order for you to guess what that material is like (except for soft is quite and hard is loud)...but that is something you have to sift through with the construction of the echo library. Serious fourier spectral analysis is involved.

The big problem with texture recognition ( I didn't even think about angle of approach!) is that a lot of the echo properties correlate to more than one physical property. Like a hard object sends a stronger echo, but so does a bigger object like a wall (although it will send back multiple echos as well, which may be the key to separating the two). A softer object sends back a weaker echo, but so does a smaller object...I can't think of a way to separate one from the other unless you used some know information about your beam-width to guess. Then again, you may not need to know if its large or small. A loud echo would mean something is worse to hit (large or hard) and a weaker echo would mean something is not so damaging to hit (soft or small)...well damaging to the robot anyways- hitting a baby is much worse than hitting an adult.

Temperature and pressure compensation is a sure thing you should do for something like that if you are into super accuracy like me. If you can figure out wind compensation (it would be far from perfect, but it would help)...multi-frequency operation or chirps would help in that situation.

It seems to you can only do the more complex stuff with widebandwidth ultrasonic transceivers.

 

[odense]

Has anybody out there played with ultrasonics for robot vision?

I'm playing with a lash-up, looking at measuring the time between sound pulse out and echo back (around 1.75mS/foot if you're interested)- nice and easy for things at right-angles to my transducers :!:
I intend to measure upto a few feet and get some sense of direction by sweeping the transducers (or a director).

There is some correllation between object size (and hardness and angle of approach) to echo strength as well- I'm considering utilising this info as possible 'texture analysis'.

If bats can do it then maybe we can?

Let look at Occupancy Grid (searching for the algorithm with US). This can help you both direction and mapping as the bat can do.

 

[odense]

you can visit website **broken link removed**
you can find some info about ultrasonic .
I also have some documents about that .If you need ,you can contact me
have fun

Hehe potay.com for non-vietnamese guys )

 

[john1]

Google doesnt seem to do Vietnamese translating.

 

[philba]

there is a project some where on the web for "ultrasonic scanning radar". I saw it about 2 years ago. The guy had mounted an untrasonic ranging unit on a servo controlled platform and scanned back and forth. He then used the range time to paint a "radar" screen. looked kind of cool. a quick google didn't turn it up so you'll have to get clever about it...

 

[john1]

radar ...

This might be the one you mean ... ?

**broken link removed**

John

 

[Nigel Goodwin]

The June 2005 issue of EPE had an 'Ultrasonic Scanner', basically an ultrasonic distance measurer mounted on a scanning system (a servo if I recall correctly?). It conected to a PC and built-up a picture on the PC screen.

 

[john1]

This one goes round and round, not back and forth,
so it may not be the one that he means.
But i would imagine its very similar.

John

 

[philba]

no, that wasn't the article I recall - it had pictures of the US scan.

 

[mvadu]

Is it this
**broken link removed**

(Sorry for repling to an old thread, i was browsing for some ultrasonic related info, i just happen to see this thread)

 

[john1]

That certainly seems like the one 'Philba' referred to.
I may be mistaken but i think it uses the guts of an 'Electronic tape
measure' the sort of thing that estate agents use for measuring to
walls.
That would be why it ignores return signals after the first one.

John