A patient fall prediction system receives video image frames from a surveillance camera positioned in a patient's room and analyses the video image frames for movement that may be a precursor to a patient fall. In set up phase, the viewpoint of the camera is directed at a risk area associated with patient falls, beds, chairs, wheelchairs, etc. A risk area is defined graphically in the viewport. The patient fall prediction system generates a plurality of concurrent motion detection zones that are situated proximate to the graphic markings of the risk areas. These motion detection zones are monitored for changes between video image frames that indicate a movement. The pattern of detections is recorded and compared to a fall movement detection signature.