Incrementally Classifying Different Walking Activities Based on Wearable Sensors

Walking environment changes in daily life, and the classification system should be able to adapt to different walking activities. In this sense, a class incremental learning method for classifying different walking activities is proposed. To demonstrate the effectiveness and performance of the proposed method, experiments are conducted with three healthy subjects. Two inertial measurement units (IMUs) and one pressure insole are used to collect the kinetic information and foot pressure of the subject, respectively. Three walking activities (six situations are included when the first walking activity is not the same) are considered. The mean classification accuracy for each walking activity is more than 93.5% at all situations. Recognition delay exists in walking transition and the largest mean delay time is 550 ms for all subjects. In addition, the recognition performance of the proposed class incremental learning method is competitive and even better than the performance of the existed recognition method that are not capable of incremental class learning.