Biometric Gait Analysis Using Wrist-Mounted Wearable Sensors

The paper presents a proposed gait biometrics system using wearable sensors. The work carried out verified the possibility of building systems using motion sensors located on the right and left wrist. The biometric system presented used input data in the form of accelerometer and gyroscope measurement values. The classifier was trained with fragments of a time series known as gait cycles - periods of time between which the participant touched his right foot against the ground. A CNN classifier with a multi-input architecture was used to validate the proposed approach. Experiments were conducted using the author’s 100-person human gait database. The results of the experiments show that the system based on the sensor located on the right wrist achieved the highest metric of 0.750 ± 0.012 F1-score, while the left wrist sensor reached 0.571 ± 0.030 F1-score.

In addition, the presented approach includes a data mechanism that increased the performance of the right wrist biometric system to 0.92 ± 0.050 and the left wrist to 0.81 ± 0.030 F1-score metrics. As a result of the augmentation experiments, it was observed that for the right and left wrist, signal perturbations should follow a different parameter selection. For the right wrist, we observed a major advantage in modeling greater tilts during movement and higher sensor vibrations. According to the literature, most people (72%) have a right dominant hand. It can be concluded that this limb is more expressive during movement and thus has greater biometric information.