What's going on? Discovering Spatio-Temporal Dependencies in Dynamic Scenes
Daniel Kuettel, Michael Breitenstein, Luc Van Gool and Vittorio Ferrari
In: CVPR 2010, 13 June - 18 June 2010, San Francisco.
We present two novel methods to automatically learn
spatio-temporal dependencies of moving agents in complex
dynamic scenes. They allow to discover temporal rules,
such as the right of way between different lanes or typical trafﬁc light sequences. To extract them, sequences of
activities need to be learned. While the ﬁrst method extracts rules based on a learned topic model, the second
model called DDP-HMM jointly learns co-occurring activities and their time dependencies. To this end we employ Dependent Dirichlet Processes to learn an arbitrary number
of inﬁnite Hidden Markov Models. In contrast to previous
work, we build on state-of-the-art topic models that allow
to automatically infer all parameters such as the optimal
number of HMMs necessary to explain the rules governing
a scene. The models are trained ofﬂine by Gibbs Sampling
using unlabeled training data.