Reduction of BRL/CAD Models and Their Use in Automatic Target Recognition Algorithms

We are currently developing an Automatic Target Recognition (ATR) algorithm to locate an object using multisensor data. The ATR algorithm will determine corresponding points between a range (LADAR) image, a color (CCD) image, a thermal (FLIR) image and a BRL/CAD model of the object being located. The success of this process depends in part on which features can be automatically extracted from the model database. The BRL/CAD models we have for this process contain more detail than can be productively used by our ATR algorithm and must be reduced to a more appropriate form. This paper presents algorithms we are developing in order to reduce the BRL/CAD models a level of detail appropriate for the ATR algorithm. A secondary feature of these algorithms is to also maintain a parallel version with details sufficient to appear realistic when rendered. This rendering enables the ATR system to animate its search procedure for monitoring and debugging. Model reduction begins by converting the Constructive Solid Geometry BRL/CAD model to a set of polyhedra. All polyhedra that are completely contained within other polyhedra are automatically discarded. We will also allow the user to discard objects based on a specified minimum bounding box. From the reduced model, we can extract 3D point sampled surface information as well as 3D features generated by the model silhouette. These features are then used by the ATR algorithm. Algorithms for determining the visible silhouette surfaces and the features for an arbitrary viewpoint are developed in this paper. The ATR algorithm will use the three dimensional features to refine a pose estimate of the model relative to sensors. This pose in turn provides a basis for measuring the quality of the match between 3D model features and sensor features. This work was sponsored by the Advanced Research Projects Agency (ARPA) under grant DAAH04-93-G-422, monitored by the U. S. Army Research Office.