Vision-Based Navigation and Recognition

Abstract : The research summarized in this report deals with many aspects of computer vision as applied to both navigation and object recognition. In particular, this research has concentrated on eight areas: parallel algorithms for vision; diffusion processes and their roles in early vision; invariant properties and their roles in object recognition; recovery of three-dimensional scene properties from single images; recovery of observer motion and scene structure from image sequences; direct motion analysis; visual interception; and vision-based navigation. Specific topics include algorithms for image and graph computations, parallel search and stereo matching; the application of diffusion processes to image morphing and face recognition; projective, affine, and deformation invariants of images; reliability of geometric computations on images of three-dimensional scenes; properties of foliage regarded as a three- dimensional texture; monocular and binocular recovery of motion and structure from feature correspondents in an image sequence; a unified treatment of feature-based and flow-based motion estimation; motion analysis based on global properties of the flow field; vision-based target interception; visibility on terrain; and landmark-based localization.