Hi James, Thanks for the suggestion! This is not something I plan to pursue right now (for reasons of cost and focus), but I will follow the links you suggested and do some more research.
As I said, we are now developing a kind of simplified visual system for Novamente that works in a simulation world where the visual sense perceives polygons. So I want to understand better what would be involved in creating a "lower level visual cortex" to feed polygons into this sort of visual system based on real-world data. But for now we have plenty of work to do with the simulation world, without worrying about real cameras... -- Ben > -----Original Message----- > From: James Anderson [mailto:[EMAIL PROTECTED] > Sent: Tuesday, October 18, 2005 11:40 AM > To: [email protected] > Subject: RE: [agi] Image to wireframe model conversion > > > In response to Ben Goertzel's question about obtaining 3D data by > an optical > method and processing it in an artificial "visual cortex." > > Here is a web site of suppliers of devices that measure 3D data. > > http://www.simple3d.com/ > > Polhemus is widely recognised as the market leader in position sensing so, > if you have $20K to burn, check out http://www.polhemus.com/fastscan.htm > > Alternatively "structured light" is a cheaper option than "laser > scanning". > Google also gives good hits with "3D scanners." However, I have no > particular advice to give about which device would be most > suitable for you. > > If you do buy a device you are certainly going to want integrated software > that returns a triangular mesh for the measured surface. But if > you want to > connect meshes up into 2D manifolds to model the 2D surface of a > 3D object, > then you are going to have to work quite hard at joining all the > surfaces up > so that there are no gaps. You will also have to cut out self-intersecting > parts of the mesh. All of this is quite a lot of topology for a programmer > or a "visual cortex" to do. But there is an easier way. > > I am a fan of simplifying things. I would supply each triangle of the mesh > with a vertex in depth, making a model using individual tetrahedra. Just > adding tetrahedra together, even with gaps and self-intersections, still > gives a topological valid result with no processing. It looks > good too, but > risks using far more tetrahedra than a manifold. > > If you think that the human brain uses masses of memory and simple > computations then my suggestion might put you on a good path. It will > certainly save you many sleepless nights of programming. > > I hope this helps, and I am sorry I was too busy to answer earlier. > > James Anderson > > ------- > To unsubscribe, change your address, or temporarily deactivate > your subscription, > please go to http://v2.listbox.com/member/[EMAIL PROTECTED] > > ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
