Activity Recognition for the Smart Hospital

A main challenge in developing ambient-intelligence environments is context recognition. Systems can easily recognize some contextual variables, such as location, but activity recognition is much more complex. The authors describe an approach for estimating activities in a working environment. They used data representing 196 hours of detailed observation of hospital workers to train and test a hidden Markov model. The results indicate that an HMM can correctly estimate user activity 92.6 percent of the time and can outperform neuronal networks and even human observers familiar with the work practices. These results can help augment ambient-intelligence environments in hospitals. This article is part of a special issue on ambient intelligence.