Axon and video camera

[Kylepowers]

I was wondering if the axon can take video in from a digital camcorder? If so how would you start it?. Im trying to not have to buy camera for my robots vision because they are expensive. If you know of ceap ones please let me know. I think 200 for one is to much. Also is it possible to build a camera that you can use with the axon and roborealm? By which i mean take a camera from a phone and make a board to connect it to and then use it with the software and hardware above. Like i would with a digital camcorder.

Thanks

[vinniewryan]

how does 10 bucks sound? 30 frames per second @ 640X480 or 15Fps @ 1300X1040
http://www.sparkfun.com/search/results?term=toshiba+camera&what=products

Although working with these things can be pretty advanced, depending on what you want to do with them.

[Kylepowers]

Are there boards for these cameras? Im trying to figure out how i would mount it. This would work perfectly if i can find the breakout board for it. Could show me were to get one?

[Kylepowers]

I also see this one hear: http://www.sparkfun.com/products/8739
Dose anyone know how to connect it to the axon?

[Soeren]

Hi,

I was wondering if the axon can take video in from a digital camcorder?

Try some math...
Eg. 640 x 480 x 15 fps = 4,608,000 bps (4.6 Mb/s plus some overhead) and that's just the raw pixel data that you need to transfer.
And then you want to apply vision algorithms to the data...

Im trying to not have to buy camera for my robots vision because they are expensive. If you know of ceap ones please let me know. I think 200 for one is to much.

Considering the amount of work you'd need to do and the years of experience you'd need to make it possible, it's not all that expensive. It's not the cameras that costs, it's the development of the software for the FPGA/ARM/whatever and since few are available, they keep the price high... When everyone and his uncle sells them, they'll be much cheaper.

Also is it possible to build a camera that you can use with the axon and roborealm? By which i mean take a camera from a phone and make a board to connect it to and then use it with the software and hardware above. Like i would with a digital camcorder.

Yes, it's possible for someone with years and years of experience and lots of time at hand, but don't expect to find someone willing to throw in hundreds of free hours to save you $200 - it's a thing that some would do because they can and find it fulfilling in some way, but it will be very expensive if the time involved is included in the total.

The proper way to go at it:
Google...
Read...
Calculate...
Experiment...
(Repeat as necessary).

Even if you don't get to a working result, you'll still have learned something along the way.

[Admin]

Computer vision requires large amounts of memory and heavy processing, neither of which any ordinary microcontroller can handle. So no, you can't hook an ordinary camera up to the Axon

have a read here:
http://www.societyofrobots.com/robot_faq.shtml#camera_microcontroller

You might be able to find an original CMUcam on Ebay for cheap. Old and limited, but you get what you pay for

[Kylepowers]

I just need something simple to do see color and see objects nothing else. Just object recignition is all I want.

[rbtying]

The $200 CMUCam* / Blackfin / etc replaces a $300 computer and a $30.00 webcam - that's a big savings in both size and $$$ in exchange for a more limited processing power.

What kind of an object are you trying to recognize?  You could probably get away with a simple line-scan sensor (TSL1401 comes to mind) and an RGB light source to determine color.  This would only require 128 * ADC_resolution (=1024?) bits in memory per reading, which might actually fit into SRAM. 

[Admin]

I just need something simple to do see color and see objects nothing else. Just object recignition is all I want.

Object recognition, or object detection? Huge difference here . . .

The CMUcam3 will give you face detection, and the Blackfin will give you basic object recognition.

If you want full object or face recognition, RoboRealm with a laptop + webcam + microcontroller is your only option.

[Kylepowers]

Pretty much any object i kind of wanted to use roborealm because its quite a nice program but im also try to be as chaep as poassible so ill take what i can get for advice. Could i just get an old webcam that i then send images wirelessly from it to a pc running roborealm then send the action back to the axon which would then interpret the command and perform it?

[Admin]

Could i just get an old webcam that i then send images wirelessly from it to a pc running roborealm then send the action back to the axon which would then interpret the command and perform it?

Do you mean a wireless webcam? Those cost between $100 and $200, and there will be lag. You would also need bluetooth to transmit commands back to your robot.

You'd be better off putting a laptop on the robot with the webcam, and the actions will be sent by USB to the Axon.

You don't want an old cheap webcam, as they could be blurry, slow, and dark. I'd recommend a Logitech with "RightLight Technology". This is what I use for RoboRealm (and chatting with the ladies ).

[Kylepowers]

I dont want to have the laptop on the robot. I dont care to much about lag there is lag either way. Little lag is alright its just object recognition and some tracking i want now. If i took this.

What kind of an object are you trying to recognize?  You could probably get away with a simple line-scan sensor (TSL1401 comes to mind) and an RGB light source to determine color.  This would only require 128 * ADC_resolution (=1024?) bits in memory per reading, which might actually fit into SRAM. 

so i could get the tracking and then a webcam for the recognition with wireless connectivity with a bluetooth sending a response back could i not get the same result even though its slow it could work.

[Admin]

I dont want to have the laptop on the robot. I dont care to much about lag there is lag either way. Little lag is alright its just object recognition and some tracking i want now. If i took this.

I'd expect lag to be about 1+ seconds, a huge problem for moving objects and moving robots . . .

What you'd need is called an IP camera:
http://www.google.com/products?hl=en&safe=off&q=ip+camera&um=1&ie=UTF-8&tbo=u&tbs=shop:1&source=og&sa=X&ei=cymATefECaTh0gG4w7TrCA&ved=0CHUQrQQ&biw=1920&bih=1042

The TRENDnet Wireless-G IP Camera is used by the mechwarfare robots, and they highly recommend it for its small size and immunity to interference:
http://www.newegg.com/Product/Product.aspx?Item=N82E16881102032&nm_mc=OTC-Froogle&cm_mmc=OTC-Froogle-_-Surveillance+-+Cameras-_-Trendnet-_-81102032

[Kylepowers]

Then i could use roborealm with it to help work out what it sees. Then use bluetooth to send back the action it would need to take right. Or would i use a wireless wifi like card?

[Admin]

Then i could use roborealm with it to help work out what it sees. Then use bluetooth to send back the action it would need to take right. Or would i use a wireless wifi like card?

To transmit data back, you'd use bluetooth (as your cheapest most reliable option).

[Kylepowers]

Ok Thanks for the input. I would like to experiment with this some.

What kind of an object are you trying to recognize?  You could probably get away with a simple line-scan sensor (TSL1401 comes to mind) and an RGB light source to determine color.  This would only require 128 * ADC_resolution (=1024?) bits in memory per reading, which might actually fit into SRAM. 

The question is could this system see depth our would you add another sensor like a sonar range finder  to see depth so that you could get a crude outline and depth of an object? I might be able to add this along side the camera. Is that possible or im i going a little far?

[rbtying]

If you want depth, use a laser pointer (you can use basic trig to find depth with a known offset), or use a Kinect (which requires a laptop on the bot). 

Sonar will give you the closest depth visible within the image, give or take, due to its rather large field of view.

[Kylepowers]

Could I use both the camera for basic shape and color. While the laser gets the deffinet shape and color along with depth or is it over kill? Which one one is cheaper and and more benificial?

 Now keep in mind I want to use a pc to take in data wirelessly from the robot and send commands to it wirelessly for higher functions like object manipulation and recicnition. While basic stuff like avoiding obsticles is handled by the axon itself. Using simple sharp irs and sonar to detect and avoid walls and such.

[Admin]

Did you read all of this yet?
http://www.societyofrobots.com/programming_computer_vision_tutorial.shtml

[Kylepowers]

Not yet is there an area I should focus on?

[Admin]

Not yet is there an area I should focus on?

That tutorial covers the absolute basics, stuff you should know as a beginner. I'd recommend at least Parts 1 to 3. Part 4 is for moving stuff, of which you don't seem to need for your project.

[Kylepowers]

Ok ill ceck it out and ill post any ideas then hear if I come up with any.

[Kylepowers]

That was a very good reading. It got me thinking i really dont need to use a camera necessarily to make this robot see. My guess is i could use a small group of sensors to just track the basic shapes and colors of the objects. What do you guys think would be the best sensors for this task?

[Admin]

For basic shapes and objects, I'd go with the Blackfin

Plus, it's WebbotLib compatible, and you get to use all my ERP source code, too.

[Kylepowers]

Ok what parts do you recommend buying? Also do you know if any where sells it cheaper possibly?

[Soeren]

Hi,

To save a bit and make it possible to handle mentally, perhaps break out a camera chip from a mouse and start playing with that (both the needed optics and the read-out).

They're low resolution cams (from around 16x16 px and up to ...64 px IIRC) and monochrome, but you'd be able to learn a lot from such a cam and if you have a mouse that's deemed unworthy of its original purpose, it could be free - some offices just throw out working ones if they eg. needs better quality. For experimenting, the most important isn't resolution, but availability of a datasheet.

[Kylepowers]

There are cameras in a pc mouse. I know theres is a laser but how would you be able to use of these cameras to do what we are talking about please inliten me. Another thought i had is what about a camera from a cellphone would that work as well?

[teoxan]

I recently bought this : http://www.sparkfun.com/products/10061, it has TTL signal and I tested it with an Arduino code.

It worked well, of course all I was receiving was bytes, but with openCV I think it will do some basic work.

It doesn't have much of a resolution but it's easy to interface it with a microcontroller and there's lot of code out there specifically for this camera.

I haven't tested it with my Axon, but I will in the following days and will let you know about.

theo

[Soeren]

Hi,

There are cameras in a pc mouse. I know theres is a laser

The LASER is just there as a light and some optical mice has an LED for that, but they all have cams to see how the mouse moves in relation to what you slide it over (by comparing consecutive samples).

but how would you be able to use of these cameras to do what we are talking about please inliten me.

This is where I have to let you in on a secret... It's called Google   

Here is a single example though.

Another thought i had is what about a camera from a cellphone would that work as well?

No, the whole point is to seriously lower the amount of pixels to reduce handling time (although I haven't looked into the speed of readout possible with a mouse cam).
It makes it easier in data handling as well to keep to the small format of a mouse cam, so it's perfect for learning.
Phone cams come in anything from 640x480 to around 5Mpixels.

[Kylepowers]

Thats cool im either thinking of useing the sparkfun as mentioned above or I might buy a cheap webcam to play with tell I can afford the blackfin. I might play with the mouse idea to. Ill think about it. The cheap web cam might be my best bet tell the blackfin can be purchased or if the sparkfun turns out.

If I go webcam what are your guys takes on setting it up.