This paper investigates the use of a commercial and low-cost RGB-D sensor for real-time facial expression recognition in Ambient Assisted Living Context. Since head poses and light conditions could be very different in domestic environments, the methodology used was designed to handle such situations. The implemented framework is able to classify four different categories of facial expressions: (1) happy, (2) sad, (3) fear/surprise, and (4) disgust/anger. The classification is obtained through an hybrid-based approach, by combining appearance and geometric features. The HOG feature descriptor and a group of Action Units compose the feature vector that is given as input, in the classification step, to a group of Support Vector Machines. The robustness of the approach is highlighted by the results obtained: the average accuracy for fear/surprise is the lowest with 85.2%, while happy is the facial …