Abstract
In smart home applications, accurate sensor-based human activity recognition is based on learning patterns online from collections of sequential sensor events. A more challenging problem is to discover and learn unknown activities that have not been observed or predefined. This is because in a real-world environment, it is impractical to presume that users/residents will only accomplish a set of predefined activities over a long-term period. To address the issues of classifying sequential data where there are multiple sensor-based activities which might be overlapping, we propose an Online Hidden Condition Random Field (OHCRF) using Resilient Gradient Algorithm (RGA) to recognize human activity behaviors. The discriminative nature of our OHCRF models the sequential observations of an online stream, resolving the level of biased data and over-fitting. The proposed adaptive RGA approach is used to update OHCRF's parameters for online learning. Compared with Stochastic Gradient Descent (SGD), the proposed adaptive RGA converges faster, and has an efficient and transparent adaptation process. Experimentally, we demonstrate that our proposed approach can outperform the state-of-the-art methods for sequential sensor-based activity recognition involving datasets acquired from residents in smart home test-beds.