Progress: August 2005 - September 2005
From LadypackWiki
| Table of contents |
Thursday, September 29, 2005
Read John Wolfe's MEng thesis (http://graphics.csail.mit.edu/pubs/thesis_jwolfe.pdf) (speech system). The part of interest was the speech system he had set up. Turns out that John was just using Galaxy and SpeechBuilder, which I've used before, so setting that up shouldn't be too hard, although I do remember it being quite tedious.
Read Roshan Baliga's MEng thesis (http://groups.csail.mit.edu/graphics/pubs/thesis_baliga.pdf) (laser range finder). Not really that much in there that I could use. Roshan described using the laser range finder, but didn't really have any tips on using it. It's possible that it's so simple to use that there wasn't any need to describe it in greater detail. Tomorrow, will actually try and use the thing...
Managed to get the ladybug2 working with Linux. Since the camera is IIDC 1.31 compliant, it's possible to just use libdc1394 to access the camera. Was confused at first about how to get images from all six cameras at once. Through an email exchange with the point grey people, I got the following:
If you look at the other available modes, you will see that the camera also supports Format 7 Modes 0, 1, 6 and 7. Here is a brief description of the modes: 0 - This is probably the mode that you are looking for. I believe that the maximum size is 1024 x 4608 and it is simply all six bayer tiled images, one after the other. 1 - The maximum size of this is 512x9216 - It is a series of 24 images - 4 for each of the 6 bayer tiled images - one for each channel. 6 - This is the same as 0 but is JPEG compressed in a bit of a proprietary format 7 - This is the same as 1 but is JPEG compressed in a similar format to 6. 0 and 1 will give you maximum throughputs of about 15fps - we would have to explain how to decode 6 and 7 in order to acheive the full 30Hz.
Wasn't getting very good frame rates with coriander, however. Paul at Point Grey indicates that ieee 1394b support on Linux isn't very good, will need to look into that. Also, there is no fancy Ladybug SDK for Linux, so even though we can get the raw images, there's no easy way to stitch them together. Is this a problem? Is it easier to just process the image data as one giant 1024x4608 image? What are the computational benefits of stitching the camera images together? Egomotion estimation from optical flow might be easier, but this could also be computed by some combination of optical flow measurements performed on each separate camera. I guess I should really look at the Ladybug2 SDK...
Also, in another exchange with the point grey people, I'm told that the ladybug2 accepts input voltages ranging from 8V - 32V. This means that the battery pack for the Ladybug 1 should also work if we get the right connectors. Am waiting for positive confirmation from pt grey. Stopped by RadioShack and picked up some connectors and will ask Ron Wiken tomorrow to show me how to put a cable together. If the old ladybug2 battery isn't sufficient, I'm borrowing a 21-cell NiMH battery from ORCA that pt grey has said will work.
Wednesday, September 28, 2005
Met with Charlie Kemp last week, who showed us his thesis work and shared his thoughts and code. Am now looking through his codebase to see what parts will be useful for us in collecting and analyzing data.
Have pretty much finished building a portable frame for the ladybug2. Bought a $10 backpack from Walmart and cut off the front part, leaving only the shoulder straps and backing. Used the laser cutter to cut out two sheets of acrylic, and bolted them to the backpack. The lower layer of plastic houses the image compression unit for the ladybug and a ton of cabling (power, optical, firewire). The laptop operating the ladybug will be strapped to the top level of plastic, alongside a battery pack for the ladybug. Bolted two more shoulder straps to the top level of plastic, from which the ladybug2 camera itself will be suspended, hanging against the chest.
Supplies used to construct this frame:
- 1 backpack, cut up.
- 2 sheets of 12"x18"x1/4" acrylic plastic, 1 sheet of 6"x6"x1/4" acrylic plastic
- 2 yards of cloth strap and buckles
- ~1' plastic tubing to use as spacers
- a bunch of zip ties
- a bunch of nuts and bolts
Unfortunately, it turns out that the battery pack for the original Ladybug does not fit/match the Ladybug2. I've emailed Point Grey, and they say that they do not make or sell battery units for the ladybug2. This means that we'll either have to run on tether for a while or assemble our own battery. It doesn't seem all that difficult to put a battery unit together, but I don't really know where to start.. Maybe Charlie will have some ideas.
Realized yesterday that the HP laptop and ieee 1394b card that I thought we had ordered didn't actually get ordered, and asked Bryt to order them. Hopefully they'll be here soon. ieee 1394b card is more important right now, because I strapped the ieee 1394b cable into the portable frame.
Short term goals:
- Try to get the ladybug2 working with linux, for linux capture (windows confirmed OK). Would just do data capture in windows, but am much more familiar with capturing audio and serial data in Linux (for microphone and laser range data and inertial sensors once we get them). ptgrey indicates some form of ieee 1394 DCAM compatibility, so will check that out.
- Find a battery?
- Figure out charlie's code, see how much I can use.
Friday, September 16, 2005
- started reading papers linked off of Olivier's citation page (http://people.csail.mit.edu/koch/research/cites.php). most are about omnidirectional video, so started Omnidirectional Video
- sent email to keith kotay last night, but no response yet.
- went with ron upstairs to see laser cutter, but machine was busy, so will try again next week.
- emailed point grey asking how to tweak ladybug2 calibration settings so that objects in the scene closer than 1 meter can be stitched together more accurately when spanning multiple cameras (between lenses).
In talking about how to carry ladybug around for collecting data, seth mentioned two possible orientations. one is frontpack, with top camera pointing forward, other is as a hat, with top camera pointing up. think i like hat idea better, because then don't 'waste' the cameras that are pointing down. seems that there really isn't much information to be obtained by looking down..
Thursday, September 15, 2005
- Picked up the ladybug2 a few days ago.
- picked up seth's thinkpad t42p to use for now until I get a laptop.
- picked up the leica laser rangefinder from the 9th floor.
- All of this is now in my office at G838.
- Have compiled a simple test program with the ladybug SDK.
- Got a copy of Olivier's code, maybe will read through it later to see what he's been doing?
Got Ron Wiken to show me the machine shop on the second floor, learned how to use all the manual machine tools except for the mill. He brought me to the laser cutter today to show me how to use that, but it was in use so I'm coming back tomorrow morning for that.
TODO
- Send an email to Keith Kotay (daniela rus's group) to ask about using the 3D printer.
- figure out how to assemble the ladybug2 into a portable unit with a microphone and the laser rangefinder. Need to design/build a chassis? seth suggested a frontpack or hat of some sorts.
- continue fiducial design. just pick something and start working with it, been doing too much background reading? The Owen, Xiao, and Middlin fiducial seems promising.
Tuesday, August 16, 2005
Larry has decided to squeeze a paper out of me for Mobile Ubiquitous Multimedia 2005. All progress slows to a crawl until Saturday, September 3 5:00 PM EDT.
Monday, August 15, 2005
bare notes from conversation with Seth:
- Fiducial Design
- Resolution analysis
- how many pixels?
- standoff
- Fiducial Appearance
- detection
- unique ID
- pose estimation
- extrinsic calibration
- Graph representation of stata?
- What can we ask or expect of user? (guided information)
- MIT naming scheme - 32-X
- CSAIL administrative naming scheme - robotics group, graphics group, architecture, etc.
- colloquial names - "Seth's room", "Larry's Room"
- gap filling (see 32-223, then an unnamed space, then 32-224, fill in the blank)
- learning adjacencies. "Learn" a graph of stata.
- inertial sensor?
a potential fiducial
+---------------+ | | | O O | | | | O | | | | | | O O | | | +---------------+
started page on Fiducial Design
