Activity Recognition in an Ambient Intelligent Environment
Ambient Intelligence (AmI):
Ambient Intelligence (AmI) is an environment in which devices are embedded and connected to each other with a communication network, working in concert to predict users’ wishes according to the context of the environment (devices and people) to help them with their everyday activities.
An ambient intelligent environment should be context-aware. One of the most complicated problems in context-aware computations is recognition of the activities in which users of the environment are engaged. These activities could be recognized by means of the information hidden in communication networks of the devices, especially different sensors embedded in the environment to ease up the process. Most of these activities usually take place sequentially in time, making hidden Markov models (HMMs) a suitable candidate to solve the problem.
In this project we investigate the problem of activity recognition in presentation ambient intelligence (PAmI). To solve this problem we have given a layered hidden Markov model which uses the data from the sensors in the first layer and passes the classification results of this layer to the second layer in order to do recognition for more complex activities taking place in a presentation.
In order to enhance the performance we need to integrate the results of recognition of different sensors into a single hidden Markov model. We have investigated different methods to do so and given a dynamic Bayesian networks (an extension of HMMs) solution for this particular problem. We have shown that this model outperforms the simple HMM. A method to combine the results of different HMMs is derived. We have shown that this method does not decrease the average error significantly but it tends to equalize the accuracy of recognition of different activities in the system.
People involved: Hamid R. Rabiee, Nasim Mirarmandehi, M. Ghazvininejad