Automatic Target Cueing Utilizing a SNAKE-Fusion Track Algorithm

Typical automatic target recognition (ATR) systems rely on measurements from images; however, acquiring the image is dependent on knowing the target location. A dynamic sensor manager points a sensor in the general target direction. Once the general target area is identified (coarse resolution), it is imperative that an ATR system increase pixels on target (fine resolution) to maintain accurate target identification. For this paper, we are concerned about maintaining target position by user-tracker reciprocal cueing. From a general wide-area search image, an operator can refine the target location by monitoring or selecting boundary points around a target. The SNAKE tracking algorithm maintains a track on a target from image sequences by developing a contour between points. For measurement drop-out, we predict target covariance from the previous image-target contour through a Kalman filter. The SNAKE-prediction region for a maneuvering target produces a precise target location from which features can be extracted for target recognition. While the SNAKE algorithm is mature, its usefulness for robust tracking is limited in that that sensor must be locked on the target for the entire process. In this development, we utilize track prediction information to follow targets through occlusions, maintain target tracks through sensor dropouts, and fuse operator inputs to refine the target location.