Online segmentation and classification of modeled actions performed in the context of unmodeled ones

In this work, we provide a discriminative framework for online simultaneous segmentation
and classification of visual actions, which deals effectively with unknown sequences
that may interrupt the known sequential patterns. To this end we employ Hough
transform to vote in a 3D space for the begin point, the end point and the label of the
segmented part of the input stream. An SVM is used to model each class and to suggest
putative labeled segments on the timeline. To identify the most plausible segments
among the putative ones we apply a dynamic programming algorithm, which maximises
an objective function for label assignment in linear time. The performance of our method
is evaluated on synthetic as well as on real data (Weizmann and Berkeley multimodal human
action database). The proposed approach is of comparable accuracy to the state
of the art for online stream segmentation and classification and performs considerably
better in the presence of previously unseen actions.