(A) causality-aware pattern mining scheme for group activity recognition in a pervasive sensor space

Abstract

Group activity recognition (GAR) is a key challenge in pervasive computing and its solutions have been presented based on various disciplines. Specifically, for GAR in a smart space without a privacy issue, data streams generated by deployed pervasive sensors are leveraged. Graphical models such as HMM and ITBM have been proposed, but their rigid structure results in a domain specificity issue. Even though emerging pattern mining is an alternative, it cannot capture an important characteristic of GAs, so-called causality. In this paper, we propose a causality-aware group activity pattern mining scheme supporting flexible representation. To preserve causally related events and extract frequent patterns of their temporal relations for each GA, we design a knowledge-based preprocessing module and pattern-tree-based pattern learning module. For evaluation, we have collected event streams generated by GAs in an IoT testbed for six months. The evaluation results show that the proposed scheme outperforms existing approaches in terms of accuracy and robustness.