Activity Recognition by Classification Method for Weight Variation Measurement with an Insole Device for Monitoring Frail People

Healthcare has become a major field of scientific research and is beginning to merge with new technologies to become connected. Measurement of motor activity provides physicians with indicators in order to improve patient follow up. One important health parameter is weight variation. Measuring these variations is not obvious when a person is walking. This paper highlights the difficulty of providing reliable weight variation values with good accuracy. To reach this objective, the paper presents ways to classify the activity of walking, in order to propose a method to measure weight variation at the right time and in a good position. Many methods were studied and compared, using Matlab. We propose a classification tree that uses the standard deviation of acceleration magnitude to define normal walking. The algorithm was embedded in an insole equipped with two force-sensing resistors and tested in laboratory.