We analyze the computational complexity per frame and
consider the two main components of the approach: the
estimation of the clusters’ boundaries over time and the cluster
assignment of the feature vectors.
Let ^ Ft = Stv
[ Btv
[ Ft, be the set of feature vectors at
frame t for t > 1, where Stv
Btv
and Ft are the Support
Vectors, Bounded Support Vectors and newly obtained feature
vectors at t, respectively. At each t, the update of Mt