Digital Signal Processing Reference
In-Depth Information
4.4.2
Surveillance Video Event Detection
In this section, we present a system for TRECVID'09 surveillance event detection
tasks. Two categories of events are detected in this system: (1) single-actor events
(i.e.,
PersonRuns
and
ElevatorNoEntry
) irrespective of any interaction between indi-
viduals, and (2) pair-activity events (i.e.,
PeopleMeet
,
PeopleSplitUp
,and
Embrace
)
that involves more than one individual. Figure
4.13
shows the framework of this
system.
The system consists of three major stages, i.e., preprocessing, event classifica-
tion, and post-processing. The preprocessing involves view classification, back-
ground subtraction, head-shoulder detection, human body detection and object
tracking. Event classification fuses One-vs.-All SVM and automata-based classifiers
to identify single-actor and pair-activity events in an ensemble way. To reduce false
alarms, events merging and post processing based on prior knowledge are applied
to refine system detection outputs.
4.4.2.1
Pair-Activity Events
Pair-activity events involve the interaction of at least two persons. This is dealt
with as a classification problem. For pair-activity events, the events of
PeopleMeet
,
PeopleSplitUp
and
Embrace
are first treated as one category and One-vs.-All SVM
is used to classify them from the others. Each kind of three events is identified by
object motion patterns.
Fig. 4.13
A system designed for TRECVID'09 surveillance event detection tasks
