Phone: (+61 8) 6488 1945
Fax: (+61 8) 6488 1089
The effect of sensor data discretisation in human activity recognition for smart spaces
Advances in smart technologies and the increasing affordability of more powerful sensing devices is bringing the future of smart spaces closer. A key requirement of this new paradigm is the detection of human activities occurring in the environment, recognised as particular combinations and patterns of events obtained from sensors, triggered by the subject's interaction with said environment.
Most current research effort in human activity recognition focuses on improving machine learning algorithms for better classification performance. The data preparation step of discretisation, also known as feature definition, selection or encoding, despite its critical impact on the quality of the generated dataset, receives inadequate attention. In my research, I investigate this fundamental problem of how to best discretise the raw sensor data of human activities to create the feature vectors for classifier training, starting with a decoupling of the different steps into separate modules.
Human activity recognition will have a significant impact in many fields, most notably at-home aged and disabled care, where remote health monitoring, emergency response and automated living assistance can decrease the burden on carers and allow people to remain in their homes for longer. In the future this will become increasingly relevant, as the proportion of older people in the population continues to rise.