|
Vision Guided Tractor
Bob H from United States [4 posts] |
11 year
|
I'm trying to visually guide a tractor up and down in a field. For now I'm not worried about turning around at the ends. Attached is a picture of my "problem space". Each "space" is 40 feet by 200 feet (the space between 2 parallel irrigation laterals.. I want to stay about 4 1/2 feet away from from the irrigation pipe and the white fence posts. I will use different width implements and one machine that drags a marker that scribes a line in the soil. I could use targets mounted on the row of trees. There is a 14 foot high fence there you can't see very well. Or I could put flags in the field like breadcrumbs to follow. Once the plants are up I could do something lane following to stay over the row. Which sequences of modules to use? How do I stay 4 1/2 feet away from the irrigation pipe and fence posts.
Can someone recommend which sequence of mods to use?
|
|
|
|
Steven Gentner from United States [1446 posts] |
11 year
|
Bob,
What you are asking for will require some environmental changes. Before we even get into the vision aspect, have you considered using high resolution gps? That's typically the tool most farmers use for guidance. Its pricey (~5K-10K) but it would give you the accuracy you need. Have a look at
http://www.hemispheregps.com/
which as that offers a global solution without needing any visual markers. They specialize in agricultural localization.
In lieu of using that technique, yes, you would ideally need some form of visual markers. It is possible to place markers along the fence but they would have to be HUGE in order to get decent position information from them. Alternatively, you can use a camera that has a really good zoom but then you'd need to know exactly what the zoom factor is in order to determine position information. For example, we are currently testing an 18ft target to determine what kind of accuracy we get using a wide angle 70deg camera. So far, it looks like at 20ft distance we get very erratic results so I don't think this will work for you.
Alternatively, local markers like colored flags or a line like you would find on a road can be used to steer the tractor. I'm not sure what possibility you have in this area. What may be worth a test is to place a camera in front of the tractor looking sideways towards the pipe and stick a checked tape ontop of the pipe. The checkered pattern is better than color since the color will change based on cloud, sunlight, etc. The alternating pattern is unique and would not exist in nature in a significant way. The angle of the camera would ensure that the tractor keeps the same distance from that line.
Something like
http://www.uline.com/Product/Detail/S-14686/Vinyl-Safety-Reflective-Tapes/2-x-10-yards-Black-White-Reflective-Tape
might work. But before you purchase large amounts its best to purchase a variety of these tapes and test to see which ones work best for you.
Naturally, this also assumes that the tape stays relatively visible. If dirt is kicked up and placed over the tape then the tracking will fail.
Another test would be to place orange/red flags on the white poles and take a couple pictures of that (with the same camera you'd use on the tractor) and see if that may work. The orange color of traffic cones normally works quite well ... maybe you can try placing a couple cones on the pipe or on the end of the poles to see what that looks like.
As you may notice, using vision requires a lot of testing and environmental changes ... that's one of the reasons that GPS is so popular in this domain.
STeven.
|
|
|
|
john wood from United Kingdom [5 posts] |
11 year
|
Could perhaps two lasers set at each end of the field with a slight angle of separation that at maximum distance they are just within maximum field of view of the camera on the tractor.
This may have to be repeated for each tractor width but with cheap pointer lasers now available would not be too expensive...that is if the beam is good enough over 200yds, perhaps near infrared with filter over the camera lens.
As it progresses up the field the both beams would eventually meet at the very end. If this is possible you could then extrapolate the position of the tractor by the beam spacing and also keep track using the beam to edge balance.............
Just an out of the box though and may be way of the mark....:-))
As for using Zoom lens, I am now on a project using an old camcorder and driving the extracted lens stepper units with a picaxe, I count the pulses used in the zoom and the math to extrapolate the measured distance on two markers ... still messing on auto focus on the lens. Will post if all turns out OK.
Cheers
John
|
|
|
|
Bob H from United States [4 posts] |
11 year
|
Thanks for the suggestions STeven.
I did think about low resolution GPS to get where I'm going with the tractor. The high rez GPS stuff is way out of my pay grade. I like the idea of cones and or tape and I might try some of that.
BTW,I noticed I can use RR to "zoom in" on a picture. Is it possible to write that into a script or program? For example: As a target gets bigger than zoom out. Repeat as necessary. Maybe at this point don't worry about an accurate distance?
∫ob
|
|
|
|
Bob H from United States [4 posts] |
11 year
|
Hi John,
A few years ago I used a cheap pen sized laser to attract a LEGO bot that just had LEDS. The activated LEDS on the bot just turned on the drive motors. But a tractor is some thing else. A friend of mine also suggested using lasers recently. I also like your idea about triangulation with lasers. So now I'm thinking some kind of flashing laser and the "laser spot module" or "laser line module"?
What kind of distance are people getting with a laser spot or vertical laser line?
My biggest question is how to make sure the camera can actually "see" the laser in the first place. In other words, how to search for and then lock on to the laser spot or vertical line?
Maybe at first just use some kind of manual control to lock on? Then begin a tracking or following senario?
Anyway thanks for all the great ideas. I'll see what works.
|
|
|
|
john wood from United Kingdom [5 posts] |
11 year
|
Hi Bob,
I was thinking off the hoof so to speak and after posting had some added ideas though the practicality is another thing..
You could have a panning laser L to R,a variable pulsed laser say 10/sec at one end increasing to 900 pulses/sec at max pan, giving 10 pulse increase /degree of rotation.
Duplicate this at the opposite corner..This allows you to accurately compute your position in the field. The above is a starting point..Use a vertical shaped lens or you could pan in the vertical so keeping maximum intensity of the laser.
The receiver could be a rotating mirror or a rotating slot with either camera or a small circle of IR light sensors..
All to do with timing but I suppose would be very accurate if the lasers are cheap and good enough for the distances...
Your tractor may end up looking like a Google Earth vehicle :-))l ..
A very interesting challenge and sure the more grey mattered members than myself will come up with a simpler answer to the problem..
Still giving it thought though .....
Good luck
John
|
|
|
|
john wood from United Kingdom [5 posts] |
11 year
|
To clarify, the lasers active on 90 deg of rotation.. tracking of the lasers would not be required, just read off the received L1 and L2 pulses will equate to your relative field position...you could go to 1/2 or even 1/4 degrees to improve accuracy if any of the suggestion is at all feasible...
|
|
|
|
Bob H from United States [4 posts] |
11 year
|
Hey John, what you suggest sounds like having 2 light houses? The kind that sit along the shore of the ocean as a reference beacon for ships? They COULD be battery powered to oscillate back and forth easy enough. Hmmmm....
Then the trick would be to come up with something on the tractor. Hmmm... maybe 2 USB cameras? one for each beacon? Each one would have it's own servo and subroutine to keep it centered on it's beacon? Each camera with some kind of encoder to record angles? Then some hocus pocus software to do all the calculations, set way points, turn the steering wheel? Hmmmm..
Maybe start with one camera and see how that goes?
I wonder if a laser could be pulsed some how so an onboard processor could determine time of flight? I'm guessing the processor would have to be pretty fast for the time of flight calculations? Not Practical? Any science or engineer type folks listening?
|
|
|
|
john wood from United Kingdom [5 posts] |
11 year
|
Hi Bob,
The pulse or modulation should not be a problem using a VCO...as the laser pans left to right covering the 90deg using a dc motor the shaft drives a pot this varying voltage is fed to a PIC that the voltage alters an oscillator in sympathy with the voltage...quite a normal set up...this then can the either switch the laser on /off,,modulate...
If the laser is also swept a few degrees up and down this would allow for the tractors bouncing up and down on uneven ground so the receiver is still swept with the lasers.
As far as the receiver goes, this would not have to track as the beams would hopefully always sweep across the sensors..you then have two beam inputs for example..at your start corner your receiver gets 10 pulses of both L1 and L2 as you move up the field you want to keep L1 as 10 pulses as L2 increases as you progress up the field which pluses relate to how far you have progressed up the field...
By drawing each lasers beam related to the field you would know where you were in the field..eg 10,10 start of tractor first run line,..10,90 at end of tractor first run line .. depending on size and shape of field you could increase accuracy by using 1/4 degrees increase frequency change.
If you look from above, in a simple graphic you would see 90 beam lines from one corner covering the field, each line as differing pulses.
Diagonally opposite you have a duplicate, as you stand anywhere in the field you would see a one off combination of pulses which directly relate to your field position.
Nothing proven here just an idea and trying just one laser would give you one axis only but would prove if it could be an answer...there is though one important thought and that is keeping the beams above eye sight level...
Cheers
John
|
|
This Thread