for you A.I. gurus

Joined
Apr 4, 2003
Messages
836
I'm looking to implement an algorithm for a project that can tell me if an object is present or not. this algorithm is fairly simple as far as vision algorithms go. all i need is a boolean condition returned to be to indicate if there is an object obstructing my vehicle's path. if there is, i need to turn. if there is not, i do not need to turn. obviously there needs to be some mechanism to ignore objects that are far away and deal with them later..

what should i research? i need some terms to look up.

thanks!

edit: i am also interested in reviewing any suggestions on how to tackle the problem without A.I. There are some constraints: it will have to be an optical means of obstruction detection, and probably will have to be infrared since it is underwater.
 
I'm looking to implement an algorithm for a project that can tell me if an object is present or not. this algorithm is fairly simple as far as vision algorithms go. all i need is a boolean condition returned to be to indicate if there is an object obstructing my vehicle's path. if there is, i need to turn. if there is not, i do not need to turn. obviously there needs to be some mechanism to ignore objects that are far away and deal with them later..

what should i research? i need some terms to look up.

thanks!

edit: i am also interested in reviewing any suggestions on how to tackle the problem without A.I. There are some constraints: it will have to be an optical means of obstruction detection, and probably will have to be infrared since it is underwater.

Well that's a dosey. What you will probably want to do is compare two images and check for signficant pixel color variation. If that exists then you have an object. The whole depth issue of 'should i leave it for later or not' becomes dicey... Something along the lines of 'if the significant pixel variations are in a sufficiently large portion of my viewing then act on it'.

Google 'robot vision' and there are plenty of freeware software parts to use...
 
What kinda hardware are you talking about using to do this? You could do this fairly easily with sonar if you wanna pay for it. Land sonar is cheap, underwater from my understanding is a lot more expensive.
 
yeah, it's pretty easy with a setup like that, but i can't use sonar. that's why i stipulated an optical requirement.

equipment would be anything otherwise necessary... cameras, IR lights, a computer, whatever it takes.... probably about a $5000 limit on parts.
 
Well that's a dosey. What you will probably want to do is compare two images and check for signficant pixel color variation.

this is the first thing that occurred to me, so it must be the simplest solution. it may work. i am concerned about the distance problem i spoke about earlier among other things. i don't want the craft to keep changing directions everytime it sees something 200 feet away.
 
How do you get the two images? How do you know which image has and which image doesn't have the obstructing object?
 
How do you get the two images? How do you know which image has and which image doesn't have the obstructing object?

assuming an IR solution for an example, an obstruction would have a large white spot. no obstruction would be significantly black.

an IR camera would need to be interfaced in this particular example of a potential solution.
 
Many auto-focus cameras use a blinking strobe light to detect and focus on objects in the dark. they also use some kind of technique to determine the closest object in a frame. Perhaps you could re-purpose a digicam to do the object detection for you?
 
this is the first thing that occurred to me, so it must be the simplest solution. it may work. i am concerned about the distance problem i spoke about earlier among other things. i don't want the craft to keep changing directions everytime it sees something 200 feet away.

Some sort of sonar ping? I saw a story a week or so ago about a guy making goggles that let him "see like a bat"... sonarish type crap.
http://www.instructables.com/id/Ultrasonic-Batgoggles/

or tack on this thing:
http://www.junun.org/MarkIII/Info.jsp?item=73
 
allow me to clarify, this vehicle is the size of a car. it is also worth reiterating that auditory guidance methods are not permitted in this environment.

thanks for the ideas though. keep 'em comin'!
 
Regarding the algorithms -- check out VisBug 1 and VisBug 2 (I think they were developed by V. Lumelsky, who currently works at NASA). You should be able to find them on scholar.google.
 
okay guys, thanks. let me address a couple suggestions.

ladar - that's essentially what i want to do, i will examine it more and see if it is an answer.

histogram - i think that was the idea modi123 came up with, and it's what i envisioned as well, but maybe modi was thinking of just comparing 2 pictures. the problem comes in that there's little indication in the histogram where the object is (which i understand isn't a requirement i told you all). it's handy to know where the object is while you approach it so that you can tell what actions you need to take. for instance, a very large object on the side in my camera view does not mean that i have to change my course, but a large object above me does because i am towing a line to the surface. i think that if i divided the pictures into several sections and did this historical analysis on them it may work better.

i will read about the VisBug algorithms as well as the other links.

thanks, all.
 
Let us know what you go with, and how it works out. This seems interesting.

Makes me wish I would have gone to grad school haha
 
actually, it's not for school. if you wanted to do something like this, it really isn't all that difficult if you're creative and have a couple friends who can be creative with you. it's all undergraduate junior-level programming, and if you couldn't figure out an algorithm, you could look it up :)

as an update, we are still evaluating options and are considering mapping the environment in 3d how a laser rangefinder would by building or converting a tool to use IR instead of laser.
 
Back
Top