State-Space based Linear Modeling for Human Activity Recognition in Smart Space

Recognition of human activity is a key element for building intelligent and pervasive environments. Inhabitants interact with several objects and devices while performing any activity. Interactive objects and devices convey information that can be essential factors for activity recognition. Using embedded sensors with devices or objects, it is possible to get object-use sequencing data. This approach does not create discomfort to the user than wearable sensors and has no impact or issue in terms of user privacy than image sensors. In this paper, we propose a linear model for activity recognition based on the state-space method. The activities and sensor data are considered as states and inputs respectively for linear modeling. The relationship between the states and inputs are defined by a coefficient matrix. This model is flexible in terms of control because all the elements are represented by matrix elements. Three real datasets are used to compare the recognition accuracy of the proposed method to those of other well-known activity recognition model to validate the proposed model. The results indicate that the proposed model achieves a significantly better recognition performance than other models.