VIDEO-ASSISTED DEEP LEARNING-BASED SMART HEALTH MONITORING SOLUTION

Abstract

With the advancement in computer-vision for healthcare, the demand for intelligent monitoring is growing rapidly. The automation in the identification of regular or irregular exercises of a patient can improve the health outcomes and can also reduce the efforts of manual monitoring. In this study, the work focuses on various health-oriented activities of a patient to calculate the range up to an abnormality scale. The proposed solution is directly dependent on the videos obtained from the ambient cameras to detect the physical movements. Modern methods are completely dependent on the amount and quality of the data used to train the system to generate an acute response to the current event. To handle these issues, a deep learning based classification methodology is proposed to address various physical abnormalities which can lead to a cause of health affliction. A subset of NTU RGB+D dataset with health-oriented examples are used to train the system. In the proposed study, 3D CNN model is used to extract the features from the videos and LSTM model is used to classify the activities.