Automatic Target Recognition (ATR) broadly refers tothe problems of target detection,acquisition/tracking. Theprocessing is completely autonomous in extracting the necessaryinformation from sensor inputs, in processing, and finaldecision-making. In addition to the traditional militaryapplications, ATR algorithms can be adopted to endow a mobilesentry robot with the capability to detect intruder generatedphenomena. Monocular parallax extracted by tracking objectsviewed from a mobile CCD camera is converted to three-dimensional(3-D) position estimates. Simulations have demonstrated that (l)the resolution and accuracy at which a robot can sample itsenvironment using a mobile CCD camera is significantly greaterthan that achieved using sonar (2) by comparing perceivedangular motion of objects in its environment with that expectedof stationary objects, the robot can detect body movement of anintruder while on the move, and (3) 3-D position estimates ofobjects in the robot's environment can be exploited forautonomous navigation, obstacle avoidance, and map generation. It is proposed that the aforementioned capability be complementedwith processing and hardware subsystems which detect differenttypes of intruder generated phenomena as follows: By comparingthe 3-D position of objects as determined on the first patrolwiththat of objects as determined on subsequent patrols, the robotcan detect a stationary intruder or disappearance of possessions. Also, microphones would allow the robot to detect an intruderwhich it can "hear" not "see."
