Comparison of orientation-independent-based-independent-based movement recognition system using classification algorithms

In the past decade, the accelerometer has been used to enable activity recognition in different application domains. In recent years, the accelerometer in a smartphone is also being applied to provide unobtrusive movement recognition. Majority of the existing investigations requires the orientation of the sensor device to be fixed. By applying the orientation-independent approach, proposed by Mizell, this requirement may be relaxed. In this paper, we compare the recognition accuracy using classification algorithms built from raw and orientation-independent acceleration data. The evaluations, based on acceleration data collected from five users, have shown that the application of the orientation-independent approach achieves accuracy up to 88 %. The trade-off of relaxing the requirement of fixed-orientation is around 5-6 %.