Active and Attentive LADAR Scanning for Automatic Target Recognition

摘要

In this work we examine the dynamic implications of active and attentive scanning for LADAR based automatic target/object recognition and show that a dynamically constrained, scanner based, ATR system\u27s ability to identify objects in real-time is improved through attentive scanning. By actively and attentively scanning only salient regions of an image at the density required for recognition, the amount of time it takes to find a target object in a random scene is reduced. A LADAR scanner\u27s attention is guided by identifying areas-of-interest using a visual saliency algorithm on electro-optical images of a scene to be scanned. Identified areas-of-interest are inspected in order of decreasing saliency by scanning the most salient area and saccading to the next most salient area until the object-of-interest is recognized. No ATR algorithms are used; instead, an object is considered to be recognized when a threshold density of pixels-on-target is reached.