Scenario, event-driven, activity recognition, formal methods

Scenarios play an important role in various computer domains. Scenarios represent activities to be described and recognized (video understanding and video surveillance, air traffic control...). Designing reliable and efficient scenario recognition tools requires to rely on a formal model. Our target scenarios should support uncertainty on input events (coming from sensors), alternative sub scenarios and different constraints (about timing, duration, ordering...). Therefore we need a model that meets all these requirements. Moreover, in order to be tracktable, the model should be hierarchical to be able to deal with smaller sub models.

The intership objective is thus first to conduct a bibliographic study about modeling techniques (e.g., stochastic grammars, hidden Markov models, probabilistic automata) and to select or design a suitable model. Since our goal is scenario recognition, the second point is to study a scenario recognition algorithm relying on the choosen model.

The student will have to achieve the following tasks:

1. Collect literature about probabilistic models leading to state of the art of underlying theories and available (real time) algorithms.
2. Propose a formal model of scenarios supporting suitable recognition algorithms.
3. Based on this model, specify a scenario description language together with its semantics; the language should be general enough to cover software engineering needs as well as reactive system ones.
4. Verify the expressiveness of the language on real examples of scenarios.

Good knowledge of object-oriented design and programming, good practice of C++ or Java, background in language theory and semantics.

PULSAR Team, INRIA-Sophia Antipolis.