Opportunistic activity recognition - pushing activity recognition  into the real world

Abstract: Activity recognition established itself as a key source of information to devise smart environments and smart assistants. Yet,current activity recognition systems are often designed to operate in well defined environment, with well defined sensor placement and modality. This limits the applicability of activity recognition in wearable computing, where the location of sensors on body and their orientation may be subject to variations, e.g. to enhance user's comfort. Nowadays, there is a wide range of sensor available in the environment. However they can't be used as-is for activity recognition, as the kind and availability of sensors changes over time depending on the location of the user and with infrastructure upgrades. Thus, the configuration of sensors available at any point in time is dynamic. Within the EU project OPPORTUNITY we devise methods to recognize human activities with "opportunistic sensor configurations", that is sensors that just happen to be available around the user, as well as on his body or sensorized objects he carries with him. We investigate frameworks, as well as new learning and adaptive techniques, to exploit sensors opportunistically discovered for activity recognition.

-> Download presentation