Titanic Smart Objects

Type: Master Thesis
Students: Andreas Breitenmoser
Advisor: Clemens Lombriser, Andreas Bulling
Project: Titan

This project presents smart objects that run Titan, the Tiny Task Network framework. Titan reprograms wireless sensor network nodes dynamically and allows the adaptation of distributed context recognition algorithms at runtime. Recognition algorithms are executed as task graphs on the nodes. Two actuator hardware modules have been developed to build the smart objects.

Based on these modules new functionality is added to Titan: the task pool is extended, a finite state machine for controlled reconfiguration is integrated and online activity recognition is performed. The activity recognition is studied with focus on efficient computation and communication to counter the resource constraints on networked embedded devices.

Demonstrator scenarios have been evaluated and the functionality of the system is demonstrated by using a small game which recognises different actions performed by the user and checks whether a number of required actions has been executed.

The project analyses recognition and discrimination of shaking movements. Recognition rates of 73.5% and 76% could be achieved with a Similarity Search classifier using an optimised set of features minimising the communication overhead for distributed recognition.