ERAR: An Event-Driven Approach for Real-Time Activity Recognition

Abstract

An exciting paradise of data is emerging into our daily life along with the development of relative perceptive technologies in smart home. How to automatically and actively recognize real-time activities from the big data is one key challenge for the future pervasive computing and ambient intelligence. Solving this problem can greatly enhance the development of relative technologies for eldercare, childcare or healthcare. This paper proposes an event-driven approach, namely activity event model, for real-time activity recognition in smart home (ERAR). The ERAR approach segments data flow based on AES Dynamic Segmentation algorithm and recognizes activities based on SVM model. The AES Dynamic Segmentation algorithm uses activity event similarity (AES) to dynamically segment data flow, and can effectively distinguish concurrent activities. Experiments in the context of smart home are presented to show that our ERAR approach performs better than the baseline approaches.