Embedded accelerometer signal normalization for cross-device gait recognition

This paper proposes a ”soft” calibration of the signal from smartphone accelerometers, with the aim to improve crossdevice gait recognition. Other applications can also benefit from the same procedure. The procedure was evaluated on a dataset of walk signals collected by three different smartphones in two timeseparated sessions. The results are extremely satisfactory. For sake of space, only the most significant ones will be reported. In some recognition settings, especially cross-device ones, a relative improvement of over 100% of the starting performance was achieved.