Human Activity Recognition with insole sensors

Human Activity Recognition (HAR) has gained a key role in the field of pervasive healthcare. The eSteps Inc. Company offers the monitoring of pre-during and post-hospitalization monitoring of motor disabilities of the lower limbs by providing intelligent insoles and help patients by defining telerehabilitation protocols. This paper offers an HAR study for detecting human activities through eSteps intelligent insoles. The study is focused on acquiring data from healthy, data cleaning and feature extraction which can help to enhance the characteristics of human activities by describing the patients’ style. Later, this paper explores the implementation of different Machine learning and Deep Learning models, according to an appropriate tuning based on Leave-One-Subject-Out Cross-Validation (LOSOCV). In real-time-scenarios, the best model reaches an accuracy of 89% and an F1-score of 90% with a temporal window of 100 samples per second.