Continuous recognition of daily activities from multiple heterogeneous sensors

Recognition of daily activities is the key to providing context-aware services in an intelligent home. This research explores the problem of activity recognition, given diverse data from multiple heterogeneous sensors and without prior knowledge about the start and end of each activity. This paper presents our approaching to continuous recognition of daily activities as a sequence labeling problem. To evaluate the capability of activity models in handling heterogeneous sensors, we compare several state-of-the-art sequence labeling algorithms including hidden Markov model (HMM), linear-chain conditional random field (LCRF) and SVM ^hmm The experimental results show that the two discriminative models, LCRF and SVM ^hmm, significantly outperform HMM. In particular, SVM ^hmm shows robustness in dealing with all sensors we used, and its recognition accuracy can be further improved by incorporating carefully designed overlapping features.