Smart mirrors are gaining attention as a smart device that could integrate a set of functionalities intended to assist older adults in their day-to-day life. These devices are seamlessly integrated in the environment, providing a user-friendly interface and naturally fitting into the daily-care routines. People face a mirror several times a day, thus ensuring that any application running on a smart mirror will have several guaranteed interactions per day. It is therefore essential to detect when the user is in front of the mirror and also to interpret what he or she is doing. Very powerful and accurate libraries are currently available, but the limited computational resources and the need to work in real time limit the valid options for smart mirror devices. This paper therefore analyses and evaluates several body pose estimation models in order to determine which one can be deployed in a smart mirror-like device dedicated to supporting older adults in their physical rehabilitation routines.