Evidential Sensor Fusion for Activity Recognition in Smart Homes
Smart Home technology provides the possibility for elderly and disabled people to stay at home independently and safely for longer periods of time. Equipped with different types of sensors a Smart Home can monitor inhabitant activities and can use sensory information to detect certain elements of change in both health and behavioural status. One of the challenges of discovering these changing elements from the sensor data is to segment sequential information perceived by sensors and merge them together to infer possible inhabitant activities.
There are different types of sensors deployed within smart environments to assist with the monitoring. The data captured by sensors may not be entirely reliable due to the environmental perturbation and the reliability of the sensors themselves. Thus reasoning inhabitant activities from uncertain, incomplete and sometimes inaccurate information is necessary whenever the monitoring system interacts with its environment. This follows directly from the fact that understanding the real world is possible only by perceiving it through a set of knowledge sources that provide partially processed sensory information.
In our current project [1], a lattice-based sensor fusion model based on the Demspter-Shafer theory of evidence along with a set of belief propagation algorithms have been developed for the purposes of inhabitant activity recognition. A preliminary evaluation has been conducted and the results attained have been promising. Although the project has promoted the application of the evidential sensor fusion approach within the Smart Home community a number of research challenges still remain.
The proposed project is aimed at further developing a framework of multi-sensor fusion by using the Dempster-Shafer theory of evidence. It will focus on three major aspects. The first one is to refine the activity recognition model developed in our ongoing work by incorporating temporal information, making the model more robust and accurate. The second is to enhance the role of statistical learning techniques in generating fusing weights which will be incorporated into the belief propagation process. The third aspect is to scale up the developed fusion model to evaluate the effectiveness of the developed model..
References:
1. Liao, J., Bi,Y., Nugent,C.: Using the Dempster–Shafer theory of evidence with a revised lattice structure for activity recognition. IEEE Transactions on Information Technology in Biomedicine, vol. 15, 2011, pp.74-82.
Personnel Involved
First Supervisor: Bi, Y Dr
Second Supervisor: Nugent, CD Professor
Collaboration: This project does not involve collaboration with another establishment