Real time activity recognition on streaming sensor data for smart environments

Real time (online) recognition of complex activities remains a challenging and active area of research. In this paper, we propose a sliding window based activity recognition (AR) method by integrating Latent Dirichlet allocation (LDA) model and Bayes theorem on real time sensor streaming. In the proposed method, we first learn offline the feature pattern of activity from activity window sequences using LDA model. We then embed a Bayes estimation of the activity probability distribution for a given sliding window in the feature extracting stage based on the learned activity-feature pattern. Finally, the probability distribution prediction as a subset of features in the sliding window is further fed into the classifier model to generate the final class result for the sliding window. We validate our approach using smart home datasets CASAS. The results of the evaluation indicate that the proposed method achieves a high accuracy of the classifier model while maintains low time cost.