The conventional K-means clustering method should
predefine the number of clusters which might be candidates
of humans or objects in our application. However, in the
service area the number of objects around the robot could not
be defined so the number of clusters should be variable. The
improved K-means clustering method [12] predefines the size
of clusters rather than the number of clusters, which makes it
possible to increase the number of clusters if the size of a
cluster is over the predefined size. Fig. 4 shows the process of
clustering.