Abstract
A Vision Based Patient Monitoring system focuses on detecting abnormal activities of a patient. In real-world, factors like occlusion and view point variations make the activity recognition task challenging. This work proposes a similarity-based representation for healthcare activities including abnormal patient activities such as coughing, sneezing, vomiting, falling, etc. Global and depth-based representations such as histogram of optical flow, displacement between skeletal sequences and relative position of skeletal joints are used to represent the spatio-temporal dynamics of activities. A benchmark data namely "NTU RGB+ D Action Recognition dataset" is used for testing the performance of the proposed approach. A comparison of the proposed methodology against other state-of-the-art approaches has proved the discrimination of the proposed approach.