Assessment of a system for gait parameter extraction and individual feature classification using artificial neural networks and a low-cost accelerometer

Abstract A system designed for monitoring the footsteps of a person is presented, aimed at determining characteristic and statistical parameters of the individual’s gait. This non-invasive approach utilizes a low-cost commercial capacitive accelerometer to sense the vibrations caused by each step as an individual walks on the floor. The system captures signals from the accelerometer, which are then processed to obtain different signal parameters (such as step duration, cadence, stride duration, kurtosis, skewness, etc), providing information about each subject under study. The collected information is stored in a database, and artificial neural networks are employed in this report to classify types or styles of walking, as well as to identify the person’s gender, age, and body mass index. With the implementation of classifiers, physical characteristics can be grouped, potentially focusing on diagnoses or identifications based on specific data. Finally, the results obtained from tests performed on 30 volunteers are presented, verifying the accelerometer’s performance and the algorithm’s effectiveness, with accuracy percentages up to 99.2% for classification. The results show a high level of coincidence and are promising for the future improvement of the system.