|
CASEE |
A representational gap exists between low-level measurements
(segmentation, object classification, tracking)
and high-level understanding of video sequences. We present a novel representation of
events in videos to bridge this gap, by extending the CASE
representation of natural languages. The proposed representation
has three significant contributions over existing
frameworks. First, it is able to represent temporal
structure and causality between sub-events. Second,
it is capable of representing multi-agent and multi-threaded
events. Last, it can detect events in an automated manner by subtree pattern matching. In order to validate our representation and automatically annotate the videos, we report experiments on several events occurring in two different domains, namely meeting video sequences, and for videos of railroad crossings.
Results for different sequences
Hand Raise (PETS video)
Stealing (USC video)
Railroad Monitoring