Abstract: A system and method are provided for automatic camera health monitoring, such as for a camera in a video surveillance system. The system preferably provides substantially continuous monitoring and detection of camera malfunction, due to either external or internal conditions. Camera malfunction is detected when a computed camera health measurement exceeds a malfunction threshold. The camera health measurement is computed based on a comparison of a current camera health record to a plurality of stored camera health records obtained in a learning mode, which characterize known states of normal camera operation. Aside from monitoring camera health, methods of the present invention can also be used to detect an object being added to or removed from a scene, or to detect a change in lighting in a scene, possibly caused by a defective light fixture.