This and the other video at vimeo.com/1153220 are the demonstration of 3D object tracking with a single webcam and Bayesian techniques.
We pick a sample of the color ball (at second 7) and by using histogram backprojection we turn the image into a probability map. Then using the Bayesian Occupancy Filter (BOF) we track 'x' and 'y', then we use information given by the BOF to estimate depth.
The 3 lines seen at the left-bottom corner are visual hints for the (x, y, z) coordinates, green is 'x', red is 'y' the cyan diagonal one is depth 'z'.
The other video ( vimeo.com/1153220 ) shows an application of this as something similar as a 3D mouse for a therapy application.
The tracking provided by BOF is very robust, it can survive partial occlusions, and when completely occluded it can restart tracking in about 2 seconds after the total occlusion is over.
Loading more stuff…
Hmm…it looks like things are taking a while to load. Try again?