Slashdot Mirror
Measure Anything with a Camera and Software
Kevin C. Tofel writes "Using a simple concept, iPhotoMEASURE software can measure any objects you can take a picture of. Include a printout of a 7.5- or 15-inch square in the photo and the software can measure any distance or object in the pic to within 99.5% accuracy. Although geared towards contractors, there's any number of consumer usage scenarios as well. Enough to justify a $99 price tag? Jury's still out on that."
...someone writes an OSS version of this? Has anyone started on this yet?
They should make a $1299 "Professional" version, quick. It'll be the exact same product as the $99 version, just in a different color, and with a printed, spiral-bound manual.
It's definitely possible to under-price your product if you're not careful. Actually, having a $1299 version might even help drive sales of the $99 version, because people would perceive the $99 version as a sort of 'deal,' as in "hey, for $99 I'm getting 60% of the features of the $1300 version! That's great! I'll take three."
"Ladies and gentlemen, my killbot features Lotus Notes and a machine gun. It is the finest available."
I was taking a mechanical design class, and I wanted to know the coordinates of a bunch of screwholes in a mounting plate. I looked at it for a second, grinned, and darted to the nearest computer with a scanner -- as my teammates shook their heads (and micrometers) at me, saying "damnit, you're being impractical; it'll never work." (They thought I was too interested in theory and not enough in turning the cranks on lathes and mills; though we generally got along, we did have -- philosophical differences.) Scanning took a few seconds, after which I took a minute to note the pixel coordinates of the hole centers in a spreadsheet. Then I measured one edge of the part with the micrometer to get a pixel-to-inch scale, popped that number into the spreadsheet, and out came the x,y coordinates of all the holes in the part. When we CNCed the new plate with those hole locations, they all lined up with the part-to-be-mounted perfectly -- at which point they were pretty much forced to admit that maybe the kid knew what the hell he was doing!
I've thought since then that some software designed for the task (with edge-recognition algorithms, measurement features, etc) could turn consumer-grade scanners into decent reverse-engineering tools (for planar parts).
It doesn't seem like it would be really too hard, if the software is just doing what I think it is.
The hardest part is just picking out the target from the photo. In most interior scenes, the target they're using would probably work pretty well (it's a white square with heavy black edges) although it seems like there are some backgrounds where locating it might be a problem. But there are, if I'm not mistaken, some OSS efforts to do things like automatic facial recognition, and that's a much more complex problem than picking a black-on-white box out. (Particularly if the center of the box is reflectorized, so that it's always 255/255/255 when a flash photo is taken.)
Once you've located the target, knowing it's actual size and how many pixels wide it is in the image, then you can let the user pick any two points elsewhere on the image (which must be in the same plane as the target, and basically perpendicular to the camera's film/sensor) and tell them how far the points are apart in reality. It's just multiplication at that point.
If you look at OSS image-processing software, there are applications around that do much more complex stuff than this: Hugin, and Panorama Tools (the latter are what really do the heavy lifting) come immediately to mind. Compared to joining and sewing a panorama, this kind of measurement seems pretty easy, unless I'm missing something critical.
If I was recommending features for a measurement product, I think the key would be not to limit it to a particular target. Sure, a few printable targets, similar to the one used in TFA's commercical product, would be good for measurement of rooms and houses, but it would also be nice to use smaller things that are typically used for scale in macro photographs. E.g., dollar bills, quarters, width of a pencil, etc. Those would be tougher to automatically recognize, and would probably require some prompting by the user in order to pick out, but would probably appeal to a wider variety of users. Who hasn't seen an eBay photo and wondered what the exact dimensions of something were?
"Ladies and gentlemen, my killbot features Lotus Notes and a machine gun. It is the finest available."
It may be possible to get ARToolkit to do this. It's dual-licensed under the GPL so it's OSS. If you want to play with it just get a v4l video capture device and print out the squares.
:) Thanks Wayne Piekarski from UniSA for letting me play ARQuake on that thing at linux.conf.au in 2003.
ARToolkit's been used by the University of South Australia to create ARQuake which is a lot of fun to use with the actual wearable computer
I'm not sure if they used ARToolkit or something more in-house to make Tinmith, that looks really sweet.
The beatings will continue until morale improves.
I used to work for one of the biggest wallpaper producers, and they were going to buy a similar program where the user prints out a sheet with some kind of arrow on it, and put it on the wall. After that, the user would take a picture of his room, and load the jpg in the application. From there, the application would calculate the angle of the wall etc, and the user could load any texture (in our case, wallpaper) and it would apply it the the user's picture of his own room. It was pretty cool, but unfortunately not web-based, it had to be given to users on CD, because it was pretty huge...and the license didn't allow just distributing it on the website..