Region-oriented Visual Attention-based Activity Detection

This paper proposes a spatio-temporal attentive mechanism
to detect events from video sequences of natural scenes of dynamic en-
vironments. More specifically, we wish to detect a visual event within a
cluttered scene, without intensive training of the algorithm. In contrast
to the event detection methods used in the literature, which drive atten-
tion based on motion and spatial location hypothesis, in our approach
the visual attention is region-driven as well as feature-driven. For this
purpose a two stages attention mechanism is proposed. In a first phase
spatio-temporal activity analysis extracts key frames from the image se-
quence and selects salient areas within these frames. For this purpose,
next to a peak detection method, we employed a change-point detection
method, which exists both in a batch as well as a incremental version.
Consequently, these areas are further processed to determine the most
interesting active region, based on a newly defined region saliency mea-
sure. The results of the proposed approach are reported using natural
image sequence of a crowded train station.