An active process for sensor-based activity data collection

The real-world activity data collection using simple and ubiquitous sensors is in general a passive process. During this process, a set of sensors are embedded with home appliances while the experience sampling tool (ESM) is provided to the user for acquiring self-reported activity label. No direct or active observation is provided to label the activity and to see the corresponding sensor activations. In such data collection, a user may end up providing wrong label or the sensor activations could be too noisy. As a consequence the resulting activity datasets are erroneous. In this paper we argue that it is important to introduce an active or direct observation for the data collection. We have introduced an intelligent tool for automatically observing the data collection process. It not only predicts the current activity of the user but also adjusts the sensor activations depending on his/her input. With the help of experimental results we have shown that the introduction of active observation can provide less noisy activity datasets.