Need ideas for a computer vision and/or machine learning application.

[slugg]

If your computer could intelligently make decisions based on images from a camera, what would you want it to do?

Some general examples:
- Face detection
- Object detection
- Horizon detection
- Spatial recognition for navigation
- Performance evaluation of mechanical processes
- Condition evaluation of materials or structures


** I need specific ideas, not general ideas. **


Here are examples of two projects I'm involved in:
- Performance Evaluation of Video Surveillance Algorithms (NIST, DoD) - completed
- Intelligence Based Image Annotation of Planetary Data Systems (NASA) - in progress

A few things that other people are doing:
- Human Emotion Detection
- Plankton Classification

Thanks and I'm interested in your feedback!

 

[CatchPhrase]

Can the photo be multiplied 60 times a second and in real time?
If so, then program it to build more of itself to manipulate and later kill humans.
I do not think they would fund that.

 

[GaryJohnson]

Performance Evaluation of Video Surveillance Algorithms (NIST, DoD) - completed

How does that work exactly? cause an algorithm is basically an evaluation, so it's evaluating an evaluation?

 

[CycloWizard]

We need an objective decision: hot or not? That's obviously the most important application you could work on.

 

[slugg]

Originally posted by: GaryJohnson

Performance Evaluation of Video Surveillance Algorithms (NIST, DoD) - completed

How does that work exactly? cause an algorithm is basically an evaluation, so it's evaluating an evaluation?

Right, based on ground-truth data. The idea was to provide a metric to track progress in the current state of the art in video surveillance algorithms in various domains. It was actually pretty cool and it "got my feet wet" in the field, so to speak

 

[slugg]

Originally posted by: CycloWizard
We need an objective decision: hot or not? That's obviously the most important application you could work on.

Funny - this could actually be done! It would be an incredibly large project in breadth, as it would involve a lot of psychology, but I think it could be possible. I have ideas on how to do it, anyhow. Nothing tested

 

[CycloWizard]

Originally posted by: slugg
Funny - this could actually be done! It would be an incredibly large project in breadth, as it would involve a lot of psychology, but I think it could be possible. I have ideas on how to do it, anyhow. Nothing tested

A lot of work has already been done in the field, though I'm not up to date on the literature so I can't say how much remains to be done. As with any field, there's always room for improvement. Go for it.

 

[slugg]

... Still seems boring to me. I'll keep it in mind though

 

[Ken g6]

I just happened to notice the link in your sig. There are lots of things biologists have to count, like endangered species. One particular example is counting fish at a dam. Do you think a computer could do this guy's job?

If that's too easy, I'm sure there are lots of "find and count $animal from aerial photography" applications out there as well.

 

[Pulsar]

How about automatic determination of potential Pornography based on the amount if surface area of skin available vs. the amount of the clothing available correcting for weight of the subject?

It'd be fun at least. I also think this one wouldn't get funded.

 

[deputc26]

I like the idea you had for spatial recognitition for navigation, but as a meteorology minor I think it would be helpful to visually analyze cloud types from terrestrial and orbital cameras and feed that data into weather forecasting models.

 

[slugg]

Originally posted by: Ken g6
I just happened to notice the link in your sig. There are lots of things biologists have to count, like endangered species. One particular example is counting fish at a dam. Do you think a computer could do this guy's job?

If that's too easy, I'm sure there are lots of "find and count $animal from aerial photography" applications out there as well.

This is a good idea, especially since they have a webcam feed. I could just come up with something on my own for fun/practice/experience.

Originally posted by: deputc26
I like the idea you had for spatial recognitition for navigation, but as a meteorology minor I think it would be helpful to visually analyze cloud types from terrestrial and orbital cameras and feed that data into weather forecasting models.

Oooo this sounds cool. Good idea. Thanks!


Anyone, please feel free to add to the list

 

[Sahakiel]

one word: Tailgate

 

[deputc26]

Here's another (though you'll be hated if you do it) airlines and other schedule based industries publish fares and schedules online. Third parties create robot programs that visit the website and copy the schedule/fares of the airline for their own use. Many airlines hate the bots slowing down their site and diluting the usability of visitor data to their site and so they implement code that interferes with the robot programs. A program that takes a "print screen style" bitmap image of the site and then uses "Machine Vision" (you could probably steal a bitmap to text engine's source code from somewhere) to extract fares would be impossible to stop without using the famously consumer hated robot stopping technology of forcing someone to extract funky text from a bitmap (maybe you could solve that too).

This would make you a soldier for the robot army you could probably make a lot of money doing it but personally I'd have some ethical issues about it.