4 ms·

"Roboticists have known for a while that ceilings are remarkably good navigation references, try as you might they are often quite different to the discerning c

"Roboticists have known for a while that ceilings are remarkably good navigation references, try as you might they are often quite different to the discerning camera." Would this work in an absolute sense, providing absolute location, or in a relative sense, like an optical mouse, sampling the ceiling and comparing each sample to the previous?

pointernil 15y ago
Why not? I guess the features would have to be large/visible enough to be picked up by the camera. For initialization some kind of qr-codes could be placed wihtin the relative positioning patterns...
stonemetal 15y ago
It ought to be good enough for absolute position. The general arrangement of lights, air vent, smoke alarm, sprinkler, motion detector, and wall proximity are more than likely unique.
- pavel_lishin 15y ago
  Assuming this technology because popular enough to affect the stores' bottom line, how much do you think an enterprising person could charge them to install a system specifically engineered to confuse this? I'm thinking, Portal-like attachments in the ceiling that randomly move around - probably at night, to avoid scaring shoppers. Or maybe you could get away with a couple of computers hooked up to half a dozen projectors, projecting random imagery on the ceiling. You could even do something interesting like a rainforest canopy, clouds, etc.
  - ChuckMcM 15y ago
    I really love this visual. Black inverted beetle hump devices, masquerading as security cameras, moving from place to place on the roof at night when the lights are out to confuse nav algorithms that defeat 'enchhanced discovery' layout.
- geon 15y ago
  If you want to make it unique, it syould be simple to cover it wit QR codes. If you know the location, size and orientation of each (which can be encoded in the QR itself), you should be able to calculate the position of the cameea pretty accurately by meassuring the perspective distortion.
jff 15y ago
If you check out SLAM positioning, you can do some really amazing stuff even if all you know is how far away from the wall you are. Given a decent floor plan, you simply take frequent distance measurements as you wander around, using dead reckoning (the accelerometer will help) to keep track of your relative positions. You rapidly build up a local map of the walls near you, which eventually will map uniquely to one location on the floor plan. Our robot only had to wander for a few seconds to figure out where it was, and that was just using a few ultrasonic rangefinders. I haven't heard of such a thing, but I imagine you could do something similarly cool with the camera of your phone. Or, even the accelerometer and compass, combined with a knowledge of the locations of doors and the available walking paths, may be able to give a reasonably good approximation of your current location--if you walked north for 30 meters, then turned left and walked another 20 meters, then north for an additional 5, in a typical mall that's going to limit the number of points you could be occupying.
- robotresearcher 15y ago
  > I haven't heard of such a thing, but I imagine you could do something similarly cool with the camera of your phone Andrew Davidson showed the first real-time monocular visual SLAM in his ICCV paper of 2003. A classic paper, and do-able on a phone these days. http://www.doc.ic.ac.uk/~ajd/Publications/davison_iccv2003.pdf http://www.doc.ic.ac.uk/~ajd/Publications/davison_iccv2003.p...

pointernil 15y ago

Why not? I guess the features would have to be large/visible enough to be picked up by the camera. For initialization some kind of qr-codes could be placed wihtin the relative positioning patterns...

stonemetal 15y ago

It ought to be good enough for absolute position. The general arrangement of lights, air vent, smoke alarm, sprinkler, motion detector, and wall proximity are more than likely unique.

pavel_lishin 15y ago
Assuming this technology because popular enough to affect the stores' bottom line, how much do you think an enterprising person could charge them to install a system specifically engineered to confuse this? I'm thinking, Portal-like attachments in the ceiling that randomly move around - probably at night, to avoid scaring shoppers. Or maybe you could get away with a couple of computers hooked up to half a dozen projectors, projecting random imagery on the ceiling. You could even do something interesting like a rainforest canopy, clouds, etc.
- ChuckMcM 15y ago
  I really love this visual. Black inverted beetle hump devices, masquerading as security cameras, moving from place to place on the roof at night when the lights are out to confuse nav algorithms that defeat 'enchhanced discovery' layout.
geon 15y ago
If you want to make it unique, it syould be simple to cover it wit QR codes. If you know the location, size and orientation of each (which can be encoded in the QR itself), you should be able to calculate the position of the cameea pretty accurately by meassuring the perspective distortion.

jff 15y ago

If you check out SLAM positioning, you can do some really amazing stuff even if all you know is how far away from the wall you are. Given a decent floor plan, you simply take frequent distance measurements as you wander around, using dead reckoning (the accelerometer will help) to keep track of your relative positions. You rapidly build up a local map of the walls near you, which eventually will map uniquely to one location on the floor plan. Our robot only had to wander for a few seconds to figure out where it was, and that was just using a few ultrasonic rangefinders. I haven't heard of such a thing, but I imagine you could do something similarly cool with the camera of your phone. Or, even the accelerometer and compass, combined with a knowledge of the locations of doors and the available walking paths, may be able to give a reasonably good approximation of your current location--if you walked north for 30 meters, then turned left and walked another 20 meters, then north for an additional 5, in a typical mall that's going to limit the number of points you could be occupying.

robotresearcher 15y ago
> I haven't heard of such a thing, but I imagine you could do something similarly cool with the camera of your phone Andrew Davidson showed the first real-time monocular visual SLAM in his ICCV paper of 2003. A classic paper, and do-able on a phone these days. http://www.doc.ic.ac.uk/~ajd/Publications/davison_iccv2003.pdf http://www.doc.ic.ac.uk/~ajd/Publications/davison_iccv2003.p...