Scenario, language theory, activity recognition

Activity recognition, for instance from video sensors, implies to recognize scenarios of interest. A scenario allows to describe a behavior: intrusion in dangerous or sensitive zones, falling down of eldery persons, vandalism ... To represent scenarios, end-users need a friendly language to describe decisive events , sub scenarios, uncertainty on input data, spatial or temporal constraints and alternative sub scenarios. Moreover, the language must be compositional to allow building complex scenarios from smaller ones. Thus, the recognition algorithm will be more efficient.

The intership objective is to design a scenario description language and to define its semantics. This semantics will interpret scenarios as terms of a probabilistic formal model. The student will develop tools to parse, verify and interpret the language together with a recognition algorithm working on scenario models.

The student will have to achieve the following tasks:

1. Study the existing formal model for scenario representation.
2. Based on this model, propose a scenario description language together with its semantics and test it on real examples of scenarios.
3. Develop tools for this language (parser, verification tools and a translator into terms of the model).
4. Propose a suitable recognition algorithm and test it in a simulated environment.

Good knowledge of object-oriented design and programming, good practice of C++ or Java, good knowledge of parsing, notions of formal methods.

PULSAR Team, INRIA-Sophia Antipolis.