robots

[yavanna]

Hi,

I am interested in an electronic guiding system for the blind and
thought it could be based on the robots' vision system
(theoretically atleast).

I wondered if robots can calculate distances between them
and an object, and if they can "see" width and depth (eg. avoid obstacle, go down the stairs,
see end of sidewalk etc...) while they are in motion?

I tried searching the net for info but all I got were catalogs for
video cams.

Thanks.

[Perspective]

computer vision is a very interesting topic in cs.

http://www-2.cs.cmu.edu/~cil/vision.html
http://iris.usc.edu/Information/Iris-Conferences.html
http://www.cs.hmc.edu/~fleck/computer-vision-handbook/

the search that started it all...
http://www.google.com/search?q=compu...utf-8&oe=utf-8

[axon]

I made a very simple light snesing robot about a year ago...it used cyonide light sensors. If you want some more info about it let me know. It was a very basic robot thought, all it did is followed a white line made out of tape on the floor, it also "saw" and avoided walls if they were white...both in dark and light areas as its path was illuminated by 8 bright white LEDs.

[major_small]

try this link:
http://www.robotstore.com
try looking up infared sensors, which a robot can use for obstacle avoidance, and line tracking, which just follows a line...

[Waldo2k2]

I think that if you were really into it, echolocation would be much more accurate. I don't think it would be any less/more difficult to program than light sensing.

[yavanna]

Thank you.

The idea for echolocation is interesting and the links
on computer vision are v. good too.

Line tracking isn't practical (unless all pavements in the world had a white line on them)
but thank you, axon, for your offer.

[major_small]

i know line tracking isn't practical in a real world, but it would work around a facility or house... in the case of sidewalks, you could use sensors on the bottom around the rim, that if they fall off the sidewalk, the robot fixes it... you also have to remember sidewalks with grass and brick sidewalks and cobblestone and street-level sidewalks, and also intersections... then you need to know wether there is a stop sign or traffic light, then you need the ability to watch for cars, or with a traffic light, watch the light and the cars...

there's a reason why no good robot has been created for these purposes...

[yavanna]

The thing is, I don't want the robot. All I want is the vision array so it could be attached to
a hand held comp and used together with some sort of voice system by a person.
(dogs can't go everywhere)

[major_small]

what exactly are you trying to create here?

what kind of visual array do you want?
distance,light,color,etc.

what do you mean by "voice system by a person"?




dogs are notoriously better at this kind of stuff than robots...

[yavanna]

Dogs are the best guids for the blind,
however there are some situations where dogs can't help.

School for example. I think it might be possible to create
a "device" to help in those circumstances.

If I use two cameras and overlap their views then I can
create a 3D image and use a computer to tell the person using
it how to avoid obstacles and hight diffrences.

It can also help them by "seeing" the blackboard for them and
store it for later refence and so on...

It might sound stupid, but I think it's doable and might not
even be expensive since the basic technology already exists.

[Waldo2k2]

>>It might sound stupid, but I think it's doable and might not
even be expensive since the basic technology already exists.

yeah it does sound stupid...whats a 3d model going to do for a blind person? you don't need to model it in 3d for the device to work, it'd actually be harder to work with a 3d rendered model than using the raw data that was used to create it.

>>It can also help them by "seeing" the blackboard for them and
store it for later refence and so on...

what good is that going to do?? they'll never be able to "see" that either...

[yavanna]

It can also help them by "seeing" the blackboard for
them and store it for later refence and so on...

what good is that going to do?? they'll never be able to "see" that either...

It will transfer it to voice data!!! They will hear it.

whats a 3d model going to do for a blind person?

I only thought about 3D for best space orientation.
I don't know if 2D is good enough.

[Waldo2k2]

>>I only thought about 3D for best space orientation.
I don't know if 2D is good enough.

you're missing the point
the data you receive is data you receive...you don't need to model it in 3d or 2d. It just has to know how far away it is from what, you can't show a computer an image and expect it to know how far away something is. You give it raw data, it uses that.

[yavanna]

I need to know not only the distance but also how wide
a thing is, and if there is a step going down or up....


Also it needs to be constantly adapting to reposition the objects
while moving.

I know a little how an infrared camera works on missile guidance
and thought to use the same principle,
but then it will be good only for space orientation and not for anything else.The same applies to echolocation.

I was thinking of a more versatile use. But I suppose I need
more technological knowlege before
I can really attack a subject like this.

Anyway, it was just an idea I had.

[Sayeh]

Determining distance with any vision system is about a thing called "triangulation'. That's why you have two eyes. It gives you depth perception because distance is calculated based on focus. When an image is brightest and edges are sharpest, the image is "in focus". Whatever mechanics it took to alter the focus so that this event would occur, are queried based on position or synchro-server feedback and those values are translated via table lookup to a "real-world" distance. It's very simple.

The 'bot' must know, by initial table data, that the motor at position "zero" means it's focused out at some specific distance (for example, an inch away). Thus, when queried, the motors can return the difference of where they are from where "zero" is, and this is translated into a focus distance.