3 ms·
In short: Compare two images with a known time difference. Remove all the parts that appear not to have moved Remove all the small stuff Comput
by RiderOfGiraffes 18y ago
In short:
Compare two images with a known time difference.
Remove all the parts that appear not to have moved
Remove all the small stuff
Compute the "centre of mass" of what's left
Do a finite difference to compute the velocity.
Repeat on another pair of images
Tracks the centroids using Alpha/Beta
(or Kalman, or unscented Kalman)
For large masses that are moving fairly consistently, this approach will work pretty well. From here you need to:
Write clearly the assumptions made
Test that your assumptions are valid
For those that aren't, decide their impact on tracking
Modify the tracking process to allow for the differences
Lather, Rinse, Repeat.
Finding and removing small centroids can be tricky, but doesn't matter so much if they're not moving. Coping with your "cloud", "rain" or "tornado" breaking up is harder. You can for example, smooth your image to help remove holes.
There's a bunch of stuff to help with that.