Detecting spontaneous collaboration in dynamic group activities from noisy individual activity data

This paper investigates the problem of recognizing activities and dynamic ad-hoc collaboration involving multiple users. Thus, we consider people performing various predominantly physical, compound activities in a smart environment (which includes personal/wearable devices). In this case, being “compound” means that the activity can be decomposed into primitive (atomic) actions that are executed by individual users. We investigate how noisy recognition of the atomic actions of individual users can be used to identify instances of cooperation at the level of the compound activities. To this end, we first introduce a hierarchical tree plan library model for activity representation. Using this new model we developed an algorithm, which allows detecting of ad-hoc team interaction without any further knowledge about roles or preliminary designed tasks. We evaluate the model and algorithm “post-mortem” with data extracted from video footage of a real nurse-emergency-training session and with increasing difficulties by artificially adding recognition-errors.