Thesis Obtaining D Silhouettes and Sampled Surfaces from Solid Models for Use in Computer Vision

OF THESIS OBTAINING D SILHOUETTES AND SAMPLED SURFACES FROM SOLID MODELS FOR USE IN COMPUTER VISION Model based object recognition algorithms identify modeled objects in a scene by relating stored geometric models to features extracted from sensor data This process can be combinatorially explosive as the amount of information presented to the recognition algorithm increases This thesis presents a method for extracting only relevant features from a stored three dimensional D model in an attempt to reduce the di culty of the recognition process The development of the methods presented here were driven by the needs of the Automatic Target Recognition ATR algorithm being developed concurrently at Colorado State University CSU The ATR algorithm locates an object using multi sensor data by determining the correspondence between a range LADAR image a color image a thermal FLIR image and a Computer Aided Design CAD geometric model The success of this process depends in part on which features can be automatically extracted from the model database Since the models available for this process contain more detail than is needed by the ATR algorithm they must rst be reduced to a more appropriate form From the reduced model we can extract D point sampled surface information as well as D model silhouette features These features are used by the ATR algorithm to re ne a pose estimate of the model relative to the sensors The pose then provides a basis for measuring the quality of the match between D model features and sensor features Mark Richard Stevens Department of Computer Science Colorado State University Fort Collins Colorado Fall