Machine Learning
IEEE Transactions on Information Technology in Biomedicine
| Hi-index | 0.00 |
In accelerometry-based activity monitoring, to correctly assess the level and recognize the type of physical activity performed by the monitored subject, it is essential to have the knowledge of the sensor wearing position. For consumer healthcare and lifestyle applications, such as with an activity monitor (AM), to restrict users with a fixed sensor position is not friendly and unrealistic as well. Therefore, it raises a need of developing a sensor position detecting method that allows a flexible sensor wearing manner and is able to extract the sensor location with very limited userdevice interactions. In this paper, two scenarios for achieving this goal are investigated. They are based on comparing the body position dependent features that are extracted from the measured acceleration data with those in an established feature database. The experiments using naturalistically collected data show the effectiveness of both scenarios, and the one that employs user-specific learning appears more promising for a practical use.