Temporal Dependency Rule Learning Based Group Activity Recognition in Smart Spaces

We present a generic framework for group activity recognition using simple non-obtrusive sensors. The proposed scheme is based on that group activity patterns can be derived from mining interval-based relationships between users' temporally overlapped actions. We leverage a hybrid architecture of probabilistic and logic knowledge that can capture the essence of the temporal dependencies, represented as a set of weighed rules. It can also learn different weights for common rules between similar group activities, which share most of sensor events and events order. The evaluation results show that our scheme outperforms the sequential baseline model, a mixture of Gaussian Hidden Markov Models.