Geometry-based Automatic Object Localization and 3-D Pose Detection

Given the image of a real-world scene and a polygonal 3-D model of a depicted object, its apparent size, image coordinates, and 3-D orientation are autonomously detected. Based on matching silhouette outline to edges in the image, an extensive search in parameter space converges to the best-matching set of parameter values. Apparent object size may a-priori be unknown, and no initial search parameter values need to be provided. Due to its high degree of parallelism, the algorithm is well suited for implementation on graphics hardware to achieve fast object recognition and 3-D pose estimation. 1