Experiments in machine vision

[Robotman]

What hardware and software is a good start for machine vision experiments?

First off, this eye ball is tethered to a PC. You can add wireless later once the concept has been proven. I think that the following would be needed:

1. A PC running Windows XP (too much hassle trying to use that "lost in space, old as the hills" Unix re-run known as Linux)

2. A video capture card with RCA inputs (even if via an adapter plug like mine has). Big Ben Computers sells the Leadtek TV2000XP for AU$65.00 - www.bigbencomputers.com.au
You could maybe buy a dedicated capture card such as the AverMedia PCI Video Capture card for AU$70.00 (also from Big Ben) because the included software might have a better API?
My [very] old WinFast TV tuner/capture card makes itself known to other video editing software under WinXP so it should be visible to Roborealm as well?

3. A video camera (cheap and cheerful) - either PAL or NTSC output. Seeing as we live in Australia then PAL ones abound. www.jaycar.com.au (CAT. NO. QC3491 [PAL] AU$49.95)
You will also need a 12V DC plug-pack too, standard center positive 2.1mm. www.jaycar.com.au (CAT. NO. MP3011 AU$17.95) The Jaycar website says it will work down to 6V but that's not true. My camera that I purchased required a minimum 8V but it sometimes resets itself. But with 9V or more, operation is perfect. Note that these cameras take about 1.5 seconds after powerup before they start outputting a signal.

4. Roborealm software - www.roborealm.com (free)

These four items will get you started. (If the Roborealm software could talk to a USB webcam then that would have been even better, but what can we do...?)

Optionally, to add some "life" to your eye ball you could get:

5. Parallax Servo Controller board (USB) - www.parallax.com/detail.asp?product_id=28823 (Stock#: 28823 US$39.95)

6. Two servo motors - one for panning and the other for tilting. Basic servos will do the job here, with a bit of [epoxy] gluing and maybe some [self-tapper] screwing to attach the camera to one servo and then that entire assemble to the other servo. Mount the entire job to some chipboard or whatever to stop it from falling over. www.jaycar.com.au (CAT. NO. YM2760 AU$24.95 times 2 = $49.90)

7. Power for the servos is 6V DC. You could use the Jaycar 1.25A REGULATED 3,4.5,6,7.5,9,12VDC SWITCHABLE SWITCHMODE POWER SUPPLY: CAT. NO. MP3031 at AU$45.00. (That's expensive - maybe you could find cheaper from a discarded toy or something.)

All these prices and sourcings are my own. I am not affiliated with any of these companies. I simply have used them in the past and find their prices and products reasonable. If you can find cheaper then please do so - I won't be offended!

So, that should be it, without any need to do any soldering at all - almost "plug and play" because you will still need to strip the wires on the 6V DC plug-pack to attach it to the servo controller board.

It almost sounds like something I could try right away. I have the PC, the servos, the camera (I have two actually), the video capture card, the software and the plug-packs, but not the servo controller board.

I will start some experiments when I have a moment of free time. Until then though, how about some of you other guys either trying this or letting me know what you think. Have I missed something?

Cheers
Botman

[Dingo]

Ok just a quick response now, more later.

Looks like a good setup although I know I'd hanker for something onboard after a short while.

Have you considered what software you want to process the signal once you've got it?
Have a look at OpenCV ( sourceforge.net/projects/opencvlibrary/ )
It can track faces and colours etc. Seems to be the de facto standard for such things and it is free.

There is a three part series on using it for beginners in SERVO at the moment - I'll see if I can get them to you.

Also what sort of application do you want the vision to do?

If you want something that can track brightly coloured objects then something like the AVRCam www.jrobot.net/Projects/AVRcam.html (or CMUCam ) will fix you up with a lot less trouble.

If you want to roll your own, CMOS 640x480 can be had for USD$20 here:
www.sparkfun.com/commerce/product_info.php?products_id=637

[Robotman]

Hi Dingo

Thanks for posting that Sparkfun camera link. This time I have made a point of reading through all of the posts in the forum to get the details. It can all be summed up nicely by visiting this webpage:

www.elinux.org/wiki/SparkfunCamera

and reading the following text dump that I made from grabbing bits from the posts:

-------------------------------------------------

Sparkfun Camera 640x480 (SKU#: SEN-00637)
www.sparkfun.com/commerce/product_info.php?products_id=637

Camera SMD Connector (SKU#: SEN-00638)
www.sparkfun.com/commerce/product_info.php?products_id=638

Camera Breakout Board (SKU#: SEN-07904)
www.sparkfun.com/commerce/product_info.php?products_id=7904

Also, www.comedia.com.hk/fp2007/comediahome.htm has camera modules.

Checkout www.elinux.org/wiki/SparkfunCamera for a wiki on this camera.

Camera Details:
---------------

3V supply maximum (2.8V is good)

The Camera echoes a clock out its PCLK port at the same freq as I run into mclk - right now, testing with 4Mhz.

It actually is an HV7131GP - not the samsung one from QIP! The pinout is identical to QIP's one [so your work wasn't in vain QIP ] - but it responds to I2C address 0x22 - which is the addy of our old friend the HV7131GP. Gonna get some better I2C test code written up after I finish this post to actually grab a register value. Then I'll be able to confirm 100%.

UPDATE: yep! reg 0 [devID] matches the HV7131GP - do you all want me to post code? its pretty ugly code .

UPDATE 2: Its definitely producing real image data - I did a quick test to see if the data is real: I stuck my multimeter on pin Y7 - which is the MSB of the luminance data. I then covered the lense - and the average voltage was ~ 0.7 V [thats an average of all the MSB's of the luminance + chrominance data - would be 0 if its just luminance]. Then I uncovered the lense - and the avg jumped to 1.4 Volts - clearly indicating more bright pixels with their MSB's set . My next target is gonna be to try and grab some image data /w my S3Board.

UPDATE 3: I've managed to get it to spit out Bayer data - no other way to get full res imagery out of it - you can either grab the raw bayer data + convert yourself - or live with YCrCb 4:2:2 which isn't as good.

I'll describe my current setup:

AVR I2C -> I2C Level Converters [just the simple fet design] -> Camera I2C
AVR Clock -> Voltage Divider [1/2] -> Camera MCLK
Camera HSYNC/VSYNC/Y* -> AVR PORTC/PORTB [IIRC]

The switching thresholds seem to be compatible with standard cmos.. - in any case - my next move is gonna be to connect this to my FPGA dev board - and try capturing the image data at higher speeds rather than the 125khz which I'm using now.

The issue isn't moving the data - its capturing it right on the negative edge.

My setup is thus:

AVR being used as an oscillator @ 4.00mhz. Fuses set for full voltage swing. One side of the xtal is being run through a voltage divider to get a 2.5V swing out of 5 Volts to clock the camera. Connected to pclk.

Output of camera [data, hsync, vsync, vclk] runs to fpga development board [Spartan 3 Starter kit from Xilinx].

logic onboard is organized as follows:
Code:

top.v - Interconnection, memory sharing, reset, etc
capture_block.v - synchronization, vsync triggering, capture on vclk edge, 4->1 cycle memory stuff

to put 4 bytes of data in a mem word, and the memory engine [just gen addr + data + strobe WE]
frequency_counter.v - basic frequency counter I use to monitor vclk when playing with I2C bits.

The rest are all the standard picoblaze files - the processor + uart + code rom.

My capture methodology goes like so:

Connect to board via hyperterm. Type 'c'. Board Confirms frame capture.
Tell hyperterm to record all data incoming via rs232.
Type 'r' - all data onboard framebuffer is streamed to computer @ 230kbaud
Save captured file - copy across network to mac where I have a decode + image display program that I've coded - just a bit of C + SDL.
Take screenshot + post for you guys!

It works fine at 4mhz - 25mhz is the rated max.

The issue is not generating the clock, its an issue with noise on the interconnects between my fpga board + the breadboard. Getting above 10mhz causes tons of crap at the clock edges, and of course on the rest of the signals.

--David Carne
-------------------------------------------------

When it comes to what I want to process the video signal for - well, you know, seeing and all that other stuff we do with our eyes. I thought that being able to see the edges of objects and know that middles of objects are solid to avoid bumping into, and maybe some recognition of some sort (faces), and overall daytime/night-time determination. Maybe even have IR LEDS to illuminate scenes when it gets dark.

The Sparkfun camera looks useful because, although David Carne uses an FPGA, it seems a basic uC will do the job if all we want is low-res greyscale. It saves buying a video ADC (such as the ADC1175) and having an LM1881 sync separator chip too. But that SMD connector - it looks really small!!! Probably want to buy the break-out board at the same time as buying the camera...

My aim is to capture the video (in low-res greyscale), pre-process it (for whatever it may need pre-processing for), and then probably build a "vector set" that defines objects (as best it can) that can be transmitted to some higher-level processing. I want to do this in stereo (first do one camera, then switch over to the other) to generate depth information. It's all experimental and still in the ideas stage at the moment.

Eventually, the system may need to break up the video into many different feature sets that each contribute a small part to the total understanding of the outside world - all linked to some sort of database that associates other cognition and thoughts and brain-like activities.

At the end of the day, the brain is one humungous processor that because of its humungous processing generates self-awareness (admittedly it would seem, to greater and lesser degrees in the general population!). It is possible for us humans to create and build our own processors that are self-aware. It is only a matter of figuring it out.

I am working on something at the moment that doesn't use image capture and is what I consider my first [very small, very limited] step towards creating this self-aware processor. This isn't "dreamworks" stuff that can never be achieved. All it takes is the doing. Putting hands in the air and saying that only God can / is allowed to do this is denying the point of why God gave us a universe full of raw materials to do things with in the first place. Crikeys, if I was God I would be actively working on creating another God (maybe that's what we are ultimately for?), so firstly I (me as God) create man and then give him the tools so that he to can create life and finally man becomes a half-decent conversation partner. Do I talk to monkeys? Heck no, so why would God want to have a chat with unthinking, unachieving followers who don't do anything except to go around saying "Holy holy holy" every nanosecond of eternity? Use your brain, use the atoms God gave us, create life, create intelligence, create.

Create.


Oh yes, that makes me feel a lot better.
Botman

[Robotman]

An update:

I've been in contact with Steven at RoboRealm who has kindly informed me that the RoboRealm software will accept a standard USB camera so there isn't any need to go and purchase a video capture card just to start basic testing.

Of course, once I want to go mobile then that's a whole other kettle of fish and will require AV transmit/receive boxes and video capture.

[donburch]

Hi Botman,

botman said:

... the RoboRealm software will accept a standard USB camera so there isn't any need to go and purchase a video capture card just to start basic testing.

Sorry I haven't been around much, or I would have pointed this out. Makes for an easy test/development rig ... even if it's only used to work out which of the OpenCV filters/algorithms to compile into your on-board controller.


... Actually I don't remember your expressing any interest in machine vision when we met. Is all this for my benefit, or for yours ? Have you decided that I'm all talk and no action, so you'll show me up ? ;D

If so .... well, thanks

[Robotman]

Dear Don

Oh, I'm interested in machine vision alright. And when you, Bones and I were having coffee we got a little disoriented once Bone's brother turned up, but I was mentioning my plans I had then for vision. The idea was to frame-slice (multiplex) two CMOS camera signals thru a single ADC1175 (ana-to-dig) chip and process each frame using the ATMEGA128 (on a JED Micro AVR570 board). Feature extraction (undecided which features at the time) and stereo depth were to be the outputs in a "high-level" coded format for later stage processing. This would be streamed via serial comms.

I've been busy lately building my Tiny Data Terminal (works very nicely too - has a numeric keypad, 2 line LCD, buzzer, and also programs 24LC256 EEPROMS). I have also built the AVR ISP Programmer kit so I can use those AT90S8535 chips I have in some robot toys I'm building. I am also self-upgrading the software that drives my "El-Cheapo" PIC programmer because Myke Predko hasn't/won't update his software to do the PIC16F88. I want to get the programmer and those toys done first, so this vision stuff is on hold for a while.

But I did get to play with the RoboRealm software the weekend before Easter. To be fair, I couldn't get anything useful out of it that I thought would lead to a machine being able to do any hand/eye coordination stuff at all. The "blobs" feature didn't work - why are there so many, and why are they all at the bottom of the screen? The ability to put boxes around things just created a zillion little boxes or one big box covering everything. And the edge/outline stuff might be useful except that I want it boxed too, but that can't be done. I played with all manner of color bit depth, grayscale (and bit depth), bright/sunlight room lighting, dark lit room, you name it. I was not satisfied. I hope you other guys have more success. And just how fast does the PC need to be? Is 2 GHz not fast enough? I found that more than three functions creates a really slow changing output.

I know there is something better out there - I just have to write it first.


So I will let you run with the vision stuff, Don. But, if you think you're not putting enough time into it, then maybe you're not putting enough time into it!!! ;D

Cheers

[donburch]

botman said:

Oh, I'm interested in machine vision alright ... I was mentioning my plans I had then for vision. The idea was to frame-slice (multiplex) two CMOS camera signals thru a single ADC1175 (ana-to-dig) chip and process each frame using the ATMEGA128 (on a JED Micro AVR570 board). Feature extraction (undecided which features at the time) and stereo depth were to be the outputs in a "high-level" coded format for later stage processing. This would be streamed via serial comms.

I continue to be amazed at how quickly what you hardware guys say goes over my head ;D

Basically you're thinking of feeding two cameras into one input stream to the ADC. Sounds to me like there will be issues synchronising the frames, and then for the ATMEGA to know whether it is looking at the Left or Right image - since it will still need to store both Left and Right images simultaneously for the depth perception comparisons.

Being a software person, I would be tempted to use 2 ADCs instead, because I'm sure my time to design/ build/ debug the frame slicing would far exceed the cost of a second ADC.

To be honest, the feature extraction and subsequent depth calculation gives me the willies 'cause I'm afraid that I won't be able to pick up the math. That's where I see RoboRealm allowing me to easily experiment with various filters. Once I know what filters and algorithms to use it should then be easier to whip up a control program, link in the appropriate subroutines from OpenCV (or similar), compile and download to the bot. I really don't like the idea of doing all the experimenting on the bot with a long program/ compile/ download/ test cycle each iteration, and with minimal debugging tools in the limited bot environment.

botman said:

But I did get to play with the RoboRealm software the weekend before Easter. To be fair, I couldn't get anything useful out of it that I thought would lead to a machine being able to do any hand/eye coordination stuff at all. The "blobs" feature didn't work - why are there so many, and why are they all at the bottom of the screen? The ability to put boxes around things just created a zillion little boxes or one big box covering everything. And the edge/outline stuff might be useful except that I want it boxed too, but that can't be done. I played with all manner of color bit depth, grayscale (and bit depth), bright/sunlight room lighting, dark lit room, you name it. I was not satisfied. I hope you other guys have more success. And just how fast does the PC need to be? Is 2 GHz not fast enough? I found that more than three functions creates a really slow changing output.

I haven't really tried RoboRealm yet, though the first few tutorials on their website (line following and object tracking) make it look very simple - as I'm sure it is when you already know which filters to combine

Their website's FAQ 12. What are the minimum requirements to run RoboRealm? recommends "CPU - As fast as possible - 2GigHz plus" with the comment that "it is very easy to quickly add a couple filters to reduce the frame rate to slower than usable".

They point out that RoboRealm runs on their BucketBot robot which is a 386, but they had to reduce the resolution to 80x60 and accept about 10fps.

botman said:

But, if you think you're not putting enough time into it, then maybe you're not putting enough time into it!!! ;D

Y'know, I just feel that I'm not putting enough time into my robot.

The base has been sitting there (well, OK now it's sitting here on my desk upstairs) for a good few months now and waiting for what ? Waiting because the spacers I picked up at Dick Smith have tiny screws which aren't long enough to go through the acrylic of Kibo's base ! Waiting three months or so for me to get around to organising some other spacers to mount the BoE controller on the acrylic. How pathetic is that

But alas that's not all. I have had my BoE-bot manual open on my desk at "Chapter 6: Light Sensitive Navigation with Photoresistors" ready ... but I haven't started the exercises. In three months I have decided to use Ultrasonics (rather than whiskers or infra-red) for collision detection, and have started to compare prices.

Sure I have been upgrading and fixing PCs (now 7 for us 3 people ); and I tell myself that I don't want to start working on something robotics in the late evening when my mind and body is already tired. So instead I busy myself with surfing the web, collecting information "to look at later".

Today I went down to Bunnings and bought a packet of 3/16" x 50mm bolts and nuts, and that job should be finished by the time I finish this damn ! 3/16 is too big to go through the mounting holes on the BoE board Maybe I'll use some wooden dowel ;D

[Robotman]

donburch said:

Basically you're thinking of feeding two cameras into one input stream to the ADC. Sounds to me like there will be issues synchronising the frames, and then for the ATMEGA to know whether it is looking at the Left or Right image - since it will still need to store both Left and Right images simultaneously for the depth perception comparisons.

Being a software person, I would be tempted to use 2 ADCs instead, because I'm sure my time to design/ build/ debug the frame slicing would far exceed the cost of a second ADC.

I would use a CD4066 to switch between one video source and the other. The ADC1175 is a relatively expensive surface-mount part. The CD4066 is cheap and has legs. The uC uses simple on/off to switch through one channel or the other. The software then has the hard job of matching features between the two captured images. Of course, the time delay between captures should be relative to a) how fast the image processing software is, and b) the maximum speed that objects in the field of view can have before the stereo processing can't do matching - ie. the object will move between one captured frame and the next, so the more change of movement (displacement) the faster the object is moving. Also, there's no software yet that also looks up a database of memorized/learned objects at the same time and does matching that way (I reckon the brain works like this at the higher level - the low level just gets the "attention circuits" happening so the eye "locks onto its target").

donburch said:

To be honest, the feature extraction and subsequent depth calculation gives me the willies 'cause I'm afraid that I won't be able to pick up the math.

I think that's where all the lack of progress has been coming from anyway. The brain/nervous system doesn't do calculus - it does phase locking and signal strength calculations, almost like a PLL radio receiver. Researchers should be letting their robots learn/be taught how to do things rather than preprogramming them. Asimo still to this day only does what its programmed to do. It can't dynamically adjust for stairs - the engineers program it to walk up and down a fixed number of stairs - add another stair and it crashes to the ground. To make it run in a circle they program it to do that - the thing is mindless, mathematically, pre-programmed. It's rubbish.

donburch said:

Sure I have been upgrading and fixing PCs (now 7 for us 3 people ); and I tell myself that I don't want to start working on something robotics in the late evening when my mind and body is already tired. So instead I busy myself with surfing the web, collecting information "to look at later".

Yep, I know what you mean. There's days when I feel the same. And the worst part is that I then feel guilty when I lock myself away from the family just to actually get something done. Having your 4 yr old sitting on your lap while trying to think and/or type at the same time is impossible. After I finished TDT (Tiny Data Terminal) I was testing it out and sending text messages to the unit and sounding its onboard buzzer and Matthew (my boy) was there with me, so I used his name and made it beep, so now everytime he comes into the workroom he says "Do my name and beep, Daddy." But of course TDT is sitting away on the shelf while I'm doing my new stuff so he gets disappointed when I say "No."

donburch said:

Today I went down to Bunnings and bought a packet of 3/16" x 50mm bolts and nuts, and that job should be finished by the time I finish this damn ! 3/16 is too big to go through the mounting holes on the BoE board Maybe I'll use some wooden dowel ;D

The problem is that you're not using a big enough hammer, Don!

Cheers!

P.S.
I didn't finish writing my own software using the El-Cheapo hardware. Instead I went and bought a PICkit2 from Farnell. (For a smidgen under A$70 I have saved myself from a job I was rapidly losing interest in.) So, tonight I will flash the latest firmware into the programmer unit and then some other night (soon!) I will build a basic ICSP (in-circuit serial programming) rig so I can make sure that all is working fine.

[Robotman]

Just an addendum to my own lack of memory:

I was watching a documentary the other night (SBS, I think) and they had a robot there that was taught the shape and sound of objects. It was clever enough to filter out the human teachers face and just concentrate on the object. If it recognized the object it would say its name (synthesized, not the humans voice simply recorded and played back), otherwise the machine would ask "What is it?" and the teacher would say the name.

I, for one, was impressed.

[donburch]

botman said:

I would use a CD4066 to switch between one video source and the other. The ADC1175 is a relatively expensive surface-mount part. The CD4066 is cheap and has legs. The uC uses simple on/off to switch through one channel or the other.

There you go again diving in to specific chips when I'm still trying to get my head around the concepts.

I don't want to get pedantic here, but years ago I was thinking of using a cheap AV switch for video editing (before I had a computer that was halfway capable), and someone pointed out that the two input sources are not co-ordinated. When you turn the knob the TV then has to re-synchronise to the start of frame from the other source, resulting in a burst of static in the output whenever you switch.

However the expensive AV switch buffers both input sources, so that the output simply reads from the other buffer at start of the next frame.

So.... I think that just arbitrarily switching between inputs may not turn out to be the best solution. Even if you detect the start of a frame and switch then, your decoder will receive a part frame which it will have to recognise and discard.

The software then has the hard job of matching features between the two captured images. Of course, the time delay between captures should be relative to a) how fast the image processing software is, and b) the maximum speed that objects in the field of view can have before the stereo processing can't do matching - ie. the object will move between one captured frame and the next, so the more change of movement (displacement) the faster the object is moving. Also, there's no software yet that also looks up a database of memorized/learned objects at the same time and does matching that way (I reckon the brain works like this at the higher level - the low level just gets the "attention circuits" happening so the eye "locks onto its target").

There's a helluva lot of what the human brain does at lower levels that we really don't understand. I was impressed by the way the human brain automagically compensates for defects in the design of the eye (in "Robot" by Rodney A Brooks).

I think that's where all the lack of progress has been coming from anyway. The brain/nervous system doesn't do calculus - it does phase locking and signal strength calculations, almost like a PLL radio receiver. Researchers should be letting their robots learn/be taught how to do things rather than preprogramming them. Asimo still to this day only does what its programmed to do. It can't dynamically adjust for stairs - the engineers program it to walk up and down a fixed number of stairs - add another stair and it crashes to the ground. To make it run in a circle they program it to do that - the thing is mindless, mathematically, pre-programmed. It's rubbish.

Certainly ASIMO is an example of simulated human bipedal movement, and NOT an example of Artificial Intelligence !

The BIG question is, how to make a robot learn something that isn't within its pre-programming - can a PLL radio receiver "learn" any other task than locking onto a radio signal ... the method of which was actually pre-programmed.

I am increasingly aware of how much our subconscious does for us ... we just think something (like recalling a humorous moment) and our subconscious automagically implements the bodily reaction (in this case a big grin ... which the female stranger on the train misinterprets and slap ). As babies we do learn how to see, to talk, to think - but as adults we no longer know how we did it.

On the other hand I feel that robotics researchers are obsessed with making machines in our own image. Are two legs really the best way to move around ? Just because dogs and spiders aren't the dominant species does not mean that 4 or 8 legs are not effective means of locomotion. For that matter human brains don't work in binary - yet that hasn't stopped researchers making very good use of the computers.

I'm interested in how my robot can achieve its objectives, and not how a human might do it. My robot uses wheels because it seems most efficient for a machine. I want to use sensors and cameras efficiently, and not be constrained by how humans work.

Also I think that we have got hung-up on the way we use computers. Sure, computer software is constantly improving - but they are small incremental steps compared with the leaps of prior decades, or the rate of hardware improvements. We are stuck in a rut and we keep trying to make computers a copy of human thinking - ignoring the fact that a computer is organised differently ... but what the new paradigm is, I cannot guess.

Yep, I know what you mean. There's days when I feel the same. And the worst part is that I then feel guilty when I lock myself away from the family just to actually get something done. Having your 4 yr old sitting on your lap while trying to think and/or type at the same time is impossible.

Since my current pay rate is low I've decided to only work 3 days ... but somehow I still don't seem to achieve more in other areas of my life, such as spending time with my 12-year-old.

donburch said:

Today I went down to Bunnings and bought a packet of 3/16" x 50mm bolts and nuts, and that job should be finished by the time I finish this damn ! 3/16 is too big to go through the mounting holes on the BoE board

The problem is that you're not using a big enough hammer, Don!

You're right there, Rod ;D

But I happened to get something for work in Jaycar last week and picked up some 1" spacers and 3mm bolts to fit, so now Kibo is in one piece. Hooray

I didn't finish writing my own software using the El-Cheapo hardware. Instead I went and bought a PICkit2 from Farnell. (For a smidgen under A$70 I have saved myself from a job I was rapidly losing interest in.)

Seems a fair price for your time

[Robotman]

donburch said:

So.... I think that just arbitrarily switching between inputs may not turn out to be the best solution. Even if you detect the start of a frame and switch then, your decoder will receive a part frame which it will have to recognise and discard.

Yep. I forgot to mention the use of an LM1881 sync separator that gives me VSYNC and HSYNC outputs so my uC can detect the start of frames, then begin storing the digitized data stream. There'll be no problem with "static" due to capturing halfway through. BTW, only black and white luminance is going to be captured. I have no intention of processing in color, yet!

donburch said:

There's a helluva lot of what the human brain does at lower levels that we really don't understand.

You are so very, very right. I didn't mean to trivialize all of the processing that does go on. From what I've read and seen, there's this amazing system that is operating all of the time, even when we are asleep, and all of the feedback paths and everything that interconnect. A recent documentary was one on synthesia and how it is revealing all of these lower-level systems.

donburch said:

The BIG question is, how to make a robot learn something that isn't within its pre-programming - can a PLL radio receiver "learn" any other task than locking onto a radio signal ... the method of which was actually pre-programmed.

Would evolution be a form of pre-programming? The neurons in the brain really do phase lock onto signals. It's all just a big jumble of pulse trains - that is, square wave signals with fixed duty cycles (pulse widths).

donburch said:

On the other hand I feel that robotics researchers are obsessed with making machines in our own image.

I've thought about that too. It's like the new paradigm that you mention (see below), in that I just can't think of a better universal physical form. Here I am admitting that I don't want to build a vertical market robot. I'm after something generic, and the human form seems (to me) the most generic and capable of all.

donburch said:

For that matter human brains don't work in binary - yet that hasn't stopped researchers making very good use of the computers.

Erm, actually our brains do work in binary. On/off pulses, remember!

donburch said:

Also I think that we have got hung-up on the way we use computers. Sure, computer software is constantly improving - but they are small incremental steps compared with the leaps of prior decades, or the rate of hardware improvements. We are stuck in a rut and we keep trying to make computers a copy of human thinking - ignoring the fact that a computer is organised differently ... but what the new paradigm is, I cannot guess.

Neither can I guess, but I know we desperately need one.

donburch said:

Since my current pay rate is low I've decided to only work 3 days ...

I heard a joke the other day that is about a boss and his worker where the punch line ends something like: "Worker: I have to work 4 days, because I can't afford more than 3 days off on what you pay me."

Cheers

[donburch]

botman said:

I heard a joke the other day that is about a boss and his worker where the punch line ends something like: "Worker: I have to work 4 days, because I can't afford more than 3 days off on what you pay me."

If you update that joke to include John Howard it'll become "I have to work 7 days because I can't afford any days off on what you pay me". For some absurd reason they have refused to look at what happened in New Zealand with similar IR laws. Probably because they are self-serving liars in the pocket of big business. Unfortunately we all get what the lowest common denominator deserve

[donburch]

I was just looking at a couple of Machine Vision articles (from Robots.net).

I guess you hardware types would be more interested in the CCD Imager project which is mostly hardware because the ATMega32 is nowhere fast enough to drive the camera.

I liked the CMUcam3 which adds an ARM7TDMI for processing the image at US$240 and uses open Source software. The technical report from the CMU Robotics Institute titled CMUcam3: An Open Programmable Embedded Vision Sensor has a reasonable amount of detail on the hardware setup. However the small on-board memory creates limitations which I consider significant (at least for the sort of things I would like to do eventually).

[ed113113]

It hadWholesale handbags been an excellentwholesale clothing pregnancy, Wholesale jewelrywithout medical wholesalerestrictions.wholesale electronics Even so,wholesale clothing he had notwholesale handbags been ableghd to make love wholesale furnitureto her forFurniture Wholesale several months.Ceramic tile He foundMicro sd cards himself wanting Wholesale clothingto protect
Wholesale Jewelry her instead,Wholesale fashion jewelry to carry her Wholesale costume jewelryup flights of stairsReplica watches, to wrap herwatches in blanketscheap watches, to bring her Replica Watchescups of custardRolex Watches. "I'm not fake rolex watchesan invalid,"replica watches she protested watcheseach time,ceramic watches laughing.omega watches

[vincent]

She wasHigh pressure blower eleven years Industrial fanyounger thanIndustrial blower he was.Commercial blower He hadwholesale first seenchina wholesale her not muchwholesale shoes more thanwholesale clothing a year ago,ghd as she rodeghd up an escalator Replica Watchesin a departmentbreitling watches store downtown,fake watch one gray November rolex watchesSaturday whileReplica Watches he waswholesale watches buying tiesRolex Watches. He was thirty-three watchesyears oldfake watch and new to Lexingtonreplica watches, Kentucky,cartier watches and she hadJewellery Watches risen out of pressure blowerthe crowd like pressure blowerssome kind offans vision,blowers her blond hair swept back in an elegant chignon, pearls glimmering at her throat and on her ears.