Robust Activity Recognition using Wearable IMU Sensors

In this paper, an orientation transformation (OT) algorithm is presented that increases the effectiveness of performing activity recognition using body sensor networks (BSNs). One of the main limitations of current recognition systems is the requirement of maintaining a known, or original, orientation of the sensor on the body. The proposed OT algorithm overcomes this limitation by transforming the sensor data into the original orientation framework such that orientation dependent recognition algorithms can still be used to perform activity recognition irrespective of sensor orientation on body. The approach is tested on an orientation dependent activity recognition system which is based on dynamic time warping (DTW). The DTW algorithm is used to detect the activities after the data is transformed by OT. The precision and recall for the activity recognition for five subjects and five movements was observed to range from 74% to 100% and from 83% to 100%, respectively. The correlation coefficient between the transformed data and the data from the original orientation is above 0.94 on axis with well-defined patterns.