Human activity recognition for physical rehabilitation using wearable sensors fusion and artificial neural networks

Physical inactivity has become one of the leading cause of death worldwide. Sedentary lifestyle is associated with an increased risk of morbidity including cardiovascular diseases, hypertension, cancer, obesity or type 2 diabetes. However, accurate exercise tracking has not been fully resolved yet. The main aim of the study was to create computable algorithm which could be implemented into wearable sensor system to help its users tracking squats as a common rehabilitation exercise. The prototype of battery-operated wearable health tracking device which tracks body temperature and body motions was developed. Pulse oximeter sensor was used to track heart rate. Seven healthy volunteers were recruited into the study. Each volunteer was asked to attach wearable device to the lower chest using elastic belt and perform activities of daily living (ADL) including sit, walk, stand and squats. All study data were recorded. The Multilayer Perceptron topology with hidden neurons was used for classification of squats. The experimental results demonstrated that the method achieved a total accuracy of 82%, sensitivity of 85% and specificity of 90%.