Two clustering algorithms and their application to motion segmentation
M. Trajkovic, Mark Hedley · 2002
Two clustering algorithms have been developed for data classification where there exists no a priori knowledge about the number of clusters. One algorithm is based on robust statistics, while the another one was motivated by the N body problem of theoretical mechanics. The algorithms were applied to the problem of motion segmentation of image sequences, based on similarity in velocity space. Experimental results for the two algorithms are presented and compared.