UMOISP: Usage Mode and Orientation Invariant Smartphone Pedometer

Abstract

Currently available pedometer applications on smartphones use either accelerometer or gyroscope sensors to calculate the number of steps walked. The main challenge with such individual sensor based approaches is that the accuracy can be impaired due to overlap in patterns when the phone is held in different modes, namely, hand (with and without swinging) and shirt pocket or pant pocket. This paper proposes a novel approach of pedometer implementation by combining both accelerometer and gyroscope sensors. By combining these two sensors and deriving features from the raw data, the proposed system can estimate step counts more accurately in all the smartphone usage modes. The proposed pedometer is also invariant to the orientation of the smartphone in each of the usage modes. This paper also proposes a novel magnetometer-based random motion detection algorithm, which can mitigate false step counts caused by random motions during phone handling. The performance of the proposed system is tested with different users across various walking conditions, and the results show an overall step count accuracy of 98.73% across all the smartphone usage modes.