A practical adaptive approach for dynamic background subtraction using an invariant colour model and object tracking

In this paper, three dynamic background subtraction algorithms for colour images are presented and compared. The performances of these algorithms defined as 'Selective Update using Temporal Averaging', 'Selective Update using Non-foreground Pixels of the Input Image' and 'Selective Update using Temporal Median' are only different for background pixels. Then using an invariant colour filter and a suitable motion tracking technique, an object-level classification is offered that recognises the behaviours of all foreground blobs. This novel approach, which selectively excludes foreground blobs from the background frames, is included in all three methods. It is shown that the 'Selective Update using Temporal Median' produces the correct background image for each input frame. The advantages of the third algorithm are: it operates in unconstrained outdoor and indoor scenes. Also it is able to handle difficult situations such as removing ghosts and including stationary objects in the background image efficiently. Meanwhile, the algorithm's parameters are computed automatically or are fixed. The efficiency of the new algorithm is confirmed by the results obtained on a number of image sequences.