Thanks, Nic! I need time now to digest what you wrote.
On Thu, 2011-03-24 at 14:00 +0200, Nic Roets wrote: > On Thu, Mar 24, 2011 at 12:08 PM, Yarik <iaroslav.shepty...@hs-bremen.de> > wrote: > > You have offered help in triangulation in one of the prev > > emails. A couple links would be awesome for a quick > > initialization. Thanks in advance! > > Firstly, sensors give you magnetism and gravity as 3D vectors. For > every picture, obtain magnetic East (in the horizontal plane) you take > the cross product of the two vectors. Then you get magnetic North by > taking the cross product of magnetic East and gravity. Then you > project the camera vector onto the horizontal plane to get the azimuth > angle. Then get elevation angle with the cosine rule. > > When the user selects multiple photos aimed at the same point, you can > estimate that point using some least squares method. Preferably the > photos should have a large angular separation (as seen from the point > being observed), otherwise the results will be inaccurate (in GPS, it > is called Dilution Of Precision). > > All these terms are described on Wikipedia, but they focus on linear > variables. Angles are non-linear. So either one should look at books > or papers describing theodolite computations, or just make your own > approximations and simulations. For example it may be sufficient to > assume the Earth is flat after applying the equirectangular > projection. > > > > > Greets, > > Yarik > > > > > > On Wed, 2011-03-23 at 20:22 +0200, Nic Roets wrote: > >> On Wed, Mar 23, 2011 at 5:30 PM, Yarik <iaroslav.shepty...@hs-bremen.de> > >> wrote: > >> > Hi Nic! > >> > > >> > Great to hear! I has been thinking of building a tool like that > >> > for quite a long time already with slightly different > >> > functionality than you describe but still very similar. I wrote > >> > already that I worked on location based mobile games. In many > >> > games players are encouraged to take pictures of different > >> > objects. The challenge I always faced in data analyzes was to > >> > extract any useful information out of those pics about features > >> > (landmarks in particular). You offer an interesting approach. Do > >> > you know anything about the last year SoC project? Would it make > >> > sense to extend its achievements? > >> > >> No, I haven't looked at it. > >> > >> > Since you say such apps are useful for community, I will develop > >> > the idea further. I will also try to apply to SoC this year. But > >> > even in case I fail I would be interested in the development. > >> > Are you going to apply for SoC too / mentoring? > >> > >> I'm not going to apply for SoC. > >> > >> I'd be happy to informally mentor anyone developing software to > >> process sensor and GPS data. I built up a little bit of experience > >> with my KeypadMapper project, as well as porting Gosmore to Android. > >> > >> In the long term (i.e. not this year) it would be nice to detect logos > >> (trademarks), perhaps do OCR or even photosynth type survey. > >> > >> But in the short term there is no image processing necessary. Simply > >> show the user a crosshair. He can then take multiple "shots" of an > >> object to collect it's height and location. > > > > > > _______________________________________________ dev mailing list dev@openstreetmap.org http://lists.openstreetmap.org/listinfo/dev
