Abstract
Objectively assessing gait function in lower limb rehabilitation remains a challenge in healthcare. This paper proposed the framework of AI gait analysis and assessment system eZiGait, which is based on seamless smart insoles. The preliminary study of activity recognition using eZiGait is presented. Walking data for five types of activities including slow walking, normal walking, fast walking, climbing upstairs, and walking down stairs have been investigated. Three classifiers were used, including artificial neural network (ANN), k-nearest neighbour (KNN) and random forest, to classify the five exercises. Results shows that a classification accuracy of 80% can be achieved with the ANN or 70% with KNN and random forest. This demonstrates that simple features extracted from smart insoles can be used to classify different types of exercise. This provides for potential development of an AI gait analysis and assistant system to support lower limb rehabilitation at hospital, community or at home using state-of-the-art smart insoles and mobile technologies.