This paper examines how a single camera can be used to determine the activities of a person or people in a smart house environment. A number of virtual sensors that correspond to various devices in the home are mapped onto the camera image such that activity detected by a virtual sensor can be used to infer interaction with that device. Such virtual sensors have similar shortcomings to other sensors in that there is confusion between actually using a device and just being near or passing the device. Image processing indoors can cause problems especially if carried out over long periods of time because of changes in the environment. For this research, background subtraction is used to detect activity along with rules to reduce false alarms. The paper will detail the map generation, image processing used and discuss an analysis of the results for data acquired in a real kitchen owned by an elderly couple. Although there is inherent ambiguity between a person actually using or just being near a device, meaningful results are obtained although much improvement in the image processing is needed.