Characterization of inertial sensors signals of a smartwatch during walking

Abstract

Abstract Parameters of the human gait cycle are commonly used for the evaluation and diagnosis of motor and neurological disorders. Traditionally, biomechanical analysis is performed in specialized laboratories, but in these locations, the subject does not perform a natural gait and tends to hide or exaggerate their alterations. Therefore, in this paper, we characterized eighteen signals per experiment: six from gyroscopes and six from accelerometers of a smartwatch (Apple Watch Series 3) worn on each wrist, and three from gyroscopes of an inertial sensors placed on the ankles. Signals were collected from twenty young and healthy subjects without pathological history. Subjects walked naturally in a straight-line for 25 meters. Each subject performed the walking cycle several times for 6 minutes, 3600 signals were thus analyzed. Each signal was characterized using time domain, frequency domain and nonlinear measurements. Results show that angular velocity in the Z-axis contains relevant information to characterize human gait in both ankles and the wrist. Also, typical gait parameters were obtained from the smartwatch signals. The sample entropy showed that signals that have greater self-similarity are those that contain more information, such as angular velocity signals on the Z-axis in each ankle and wrist. This characterization could be useful for the automatic identification of the human gait cycle, the detection of pathologies and the recognition of people from human gait patterns using only the information extracted from sensors embedded in smartphones and smartwatches.