This work deals with the problem of human activity recognition. It is greatly motivated by research in video activity understanding applied to the domain of health care monitoring. Research on activity recognition is receiving an increasing attention from the scientific community today. It is one of the most challenging problem in computer vision and artificial intelligence domains. The main goal of the current activity recognition research consists in recognizing and understanding short-term action and long-term complex activities. In this work, we propose two main contributions. The first contribution consists of an approach for video activity recognition that addresses the uncertainty management issues for accurate event detection. The second contribution consists in defining an ontology and a knowledge base for health care monitoring and in particular Alzheimer monitoring at hospital. The proposed activity recognition approach combines semantic modelling together with a probabilistic reasoning to cope with the errors of low-level detectors and to handle activity recognition uncertainty. The probabilistic recognition of activities is based on Bayesian probability theory which provides a consistent framework for dealing with uncertain knowledge. The proposed probabilistic constraint verification approach based on Gaussian probability model enforces the accuracy of the activity recognition algorithm. We work in close collaboration with clinicians to define an ontology and a knowledge base for Alzheimer monitoring at hospital. The defined ontology contains several concepts useful for health care. We also define a number of criteria which could be observed by camera sensors to allow detection of early symptoms of Alzheimer's disease. We validate the proposed algorithm on real-world videos. The experimental results show that the proposed activity recognition algorithm can successfully recognize activities with a high recognition rate. The obtained results for health care monitoring highlight the advantages of the use of the proposed approach as a support platform for clinicians to objectively measure patient performance and obtain a quantifiable assessment of instrumental activities of daily living and gait analysis.
