We propose an unsupervised learning technique for the automatic classification of activities of daily living (ADL) from video data collected in-place. This technique may be used to develop automated home care systems capable of context-aware interactions with older adults. Data was acquired from a video camera mounted on the ceiling of a simulated bedroom while a subject performed unconstrained activities relevant to this context. A constrained Gaussian model, obtained through Factor Analysis (FA), was later fitted to all frames in the video. The required number of activities to be detected corresponded to the number of latent variables in the constrained model. The mean of the posterior distribution for each previously unseen frame was calculated, and a label corresponding to the factor with the maximum posterior mean for each frame was assigned. As it was naturally expected, the accuracy of our classifier was dependent on the number of activity classes required with a maximum of up to 82.57% of true positives achieved for only two generic activity classes. Sequential arrangements of multiple classifiers may be used to increase the number of activities accurately detected. The simplicity of this method allows for the trivial implementation of real-time, context-aware and interactive home care systems. Current limitations are also discussed.