Tracking object paths across sparse array
in [Matlab]
|
Prev: Aligning Point Clouds
Next: AM
From: Steven Lord on 29 Jun 2010 13:09 "Matt J " <mattjacREMOVE(a)THISieee.spam> wrote in message news:i0d3rn$f7p$1(a)fred.mathworks.com... > "us " <us(a)neurol.unizh.ch> wrote in message > <i0d3c8$cb5$1(a)fred.mathworks.com>... > >> > >> > Then to proceed further you need to specify a mathematical description >> > for how the flight paths progress. That's the only hope of >> > distinguishing which path belongs to which plane at the crossing >> > points. >> >> now: WHAT if there is NO continuation after a crossing point(?)... > ========== > > You mean due to a mid-air collision? That would, of course, add a few > if/else branches to the code... That's one possibility -- but another possibility is that the paths of the two flights coincided for some distance so that the two flights appear as one path. On the ground, this is easy to imagine; one car drives on a highway, and later on another car enters that same highway and travels in the same lane the first car used for some distance before exiting the highway. In the air this is probably less likely, unless the flights are required to take one of a smaller number of approach paths when coming in for a landing. -- Steve Lord slord(a)mathworks.com comp.soft-sys.matlab (CSSM) FAQ: http://matlabwiki.mathworks.com/MATLAB_FAQ To contact Technical Support use the Contact Us link on http://www.mathworks.com
From: Matt J on 29 Jun 2010 13:28 "Steven Lord" <slord(a)mathworks.com> wrote in message <i0d9bs$h94$1(a)fred.mathworks.com>... > > > > You mean due to a mid-air collision? That would, of course, add a few > > if/else branches to the code... > > That's one possibility -- but another possibility is that the paths of the > two flights coincided for some distance so that the two flights appear as > one path. On the ground, this is easy to imagine; one car drives on a > highway, and later on another car enters that same highway and travels in > the same lane the first car used for some distance before exiting the > highway. In the air this is probably less likely, unless the flights are > required to take one of a smaller number of approach paths when coming in > for a landing. ==================== True. Again though, this tracking problem only seems hard if the flight paths overlap both spatially and temporally. In the approach path scenario you describe, spatial overlap seems plausible, but temporal?
From: Matt J on 29 Jun 2010 13:35 "Matt J " <mattjacREMOVE(a)THISieee.spam> wrote in message <i0daf7$4d$1(a)fred.mathworks.com>... > True. Again though, this tracking problem only seems hard if the flight paths overlap both spatially and temporally. In the approach path scenario you describe, spatial overlap seems plausible, but temporal? ================= So if this hypothesis is correct, I'm starting to think that the problem can be solved using bwconncomp after all. You could stack the 500 time frames to form a 3D logical array and use bwconncomp to segment the flight paths as disconnected curves in (x,y,t) space.
From: Dan on 29 Jun 2010 13:38 "Matt J " <mattjacREMOVE(a)THISieee.spam> wrote in message <i0daf7$4d$1(a)fred.mathworks.com>... > "Steven Lord" <slord(a)mathworks.com> wrote in message <i0d9bs$h94$1(a)fred.mathworks.com>... > > > > > > > You mean due to a mid-air collision? That would, of course, add a few > > > if/else branches to the code... > > > > That's one possibility -- but another possibility is that the paths of the > > two flights coincided for some distance so that the two flights appear as > > one path. On the ground, this is easy to imagine; one car drives on a > > highway, and later on another car enters that same highway and travels in > > the same lane the first car used for some distance before exiting the > > highway. In the air this is probably less likely, unless the flights are > > required to take one of a smaller number of approach paths when coming in > > for a landing. > ==================== > > > True. Again though, this tracking problem only seems hard if the flight paths overlap both spatially and temporally. In the approach path scenario you describe, spatial overlap seems plausible, but temporal? The array I am trying to seperate the individual plane paths from is a sparse matrix with a number other than zero whereever a plane has been, the x and y axis being the x and y coordinates. So if two planes ever passed over the same spot, it registers only once.
From: ImageAnalyst on 29 Jun 2010 13:41
On Jun 29, 12:08 pm, "Matt J " <mattjacREM...(a)THISieee.spam> wrote: > Come to think of it, you might not need a mathematical model as long as the planes never cross paths at the same (if they do, it would hopefully be at different altitudes, cf. us). > > In that case, simply track the positions from frame to frame. I.e., in each of your 500 frames, find the (x,y) positions of the planes using find(). > > Then, for each (x,y) position, find the closest one in the subsequent frame using bsxfun(). That will allow you to match planes between adjacent frames. --------------------------------------------------------------------------------------------------------- Dan: I agree with Matt J. The only way to remove ambiguity (almost completely) is to track each plane one at a time as it progresses through the frames. It would help if you could post pictures. I'm envisioning each frame being black with 10 white dots, each dot representing a plane. So just find the centroids of the dots, and in the next frame, see which centroids and trajectories makes sense when compared to the dots of the prior frames. The only possible ambiguity would come if, as Steven said, you have complete overlap for a certain trajectory (so there are only 9 dots in some pictures instead of 10). But you do need to keep track of the trajectory angle because just tracking centroids may not be enough if two of the dots get very close to each other (not even overlapping, just close). -ImageAnalyst |