Through a chain of transformation steps, we seek to obtain an efficient algorithm. Specification - Alg - ... 3 Algorithm Each transformation step preserves input-output equivalence, so the final algorithm requires no additlonal verification. Algorithm design is a difficult artificial intelligence task involving representation and planning issues. First, in reasoning about a complicated object like an algorithm it is essential to divide it into parts that interact in relatively simple ways. We have chosen asynchronous processes, communicating via data channels, as an appropriate representation for algorithms.
AN OPTIMIZATION APPROACH FOR USING CONTEXTUAL INFORMATION IN COMPUTER VISION ABSTRACT Olivier D. Faugeras Image Processing Institute University of Southern California Los Angeles, California 90007, U.S.A. Local parallel processes are a very efficient way of using contextual information in a very large class of problems commonly encountered in Computer Vision. An approach to the design and analysis of such processes based on the minimization of a global criterion by local computation is presented. INTRODUCTION The problem of assigning names or labels to a set of units/objects is central to the fields of Pattern Recognition, Scene Analysis and Artificial Intelligence. Of course, not all possible names are possible for every unit and constraints exist that limit the number of valid assignments. These constraints may be thought of as contextual information that is brought to bear on the particular problem, or more boldly as a world model to help us decide whether any particular assignment of names to units makes sense or not.
HUMAN MOVEMENT UNDERSTANDING: A VARIETY OF PERSPECTIVES Norman I. Badler Joseph O'Rourke Stephen Platt Mary Ann Morris Department of Computer and Information Science Moore School D2 University of Pennsylvania Philadelphia, PA 19104 ABSTRACT Our laboratory is examining human movement from a variety of perspectives: synthesis of animated movements and analysis of moving images. Both gestural (limb and hand) and facial movements are being explored. I HUMAN MOVEMENT SYNTHESIS Our laboratory is examining human movement representation from a variety of perspectives, including synthesis of three-dimensional animated movements and analysis of moving images. These broad areas are further refined into gestural (limb and hand) and facial movements since very different sorts of actions take place in jointed skeletal movement and "rubber-sheet" facial distortions. Our human body model r5](Figure 1) and hand model (Figure 1) are based on spherical decompositions of three-dimensional objects.
High-contrast lighting and thresholding may be used to obtain an accurate silhouette that can be processed at video rates to yield useful features, such as area, perimeter, centroid, and higher moments. In addition, structural information is available in the geometric relationships between the local features of the outline (holes, corners, and so on). This kind of information is sometimes sufficient for some industrial automation (IA) tasks, such as part identification and acquisition. Other tasks, however, are not so easily approached. Although many simple parts can be adequately represented by a mere outline, most assemblies cannot because they are typically composites of several overlapping parts or subassemblies.
INTERFERENCE DETECTION AND COLLISION AVOIDANCE AMONG THREE DIMENSIONAL OBJECTS* N. Ahuja, R. T. Chien, R. Yen, and N. Bridwell ABSTRACT Two methods for detecting intersections among three dimensional objects are described. The first method involves detecting overlap among the projections of the objects on a given set of planes. The second method uses a three dimensional octree representation of the objects. Intersections are detected by traversing the trees for the obstacles and the moving objects. Application of the methods to collision avoidance is discussed. I INTRODUCTION An important problem in the robotics manipulation of its environment is that of following collision free trajectories when objects have to be moved.
ABSTRACT process is robust, because it bases its decisions on grou s of mutual1 consistent features, and it is rela P-ively fast, % ecause it concentrates on key features that are automatically selected on the basis of a detailed analysis of CAD type of models of the objects. INTRODUCTION There are several tasks that involve locating partial1 relative 4 y visible easy tasks, objects. This approach is fast because it locates the minimum number of features; however, if the ob'ects are complicated, determining the or a er of the features to be located may be difvficult. Development of the location strategy becomes even more difficult when mistakes are taken into account. In the parallel approach, all the features in an image are located, and then large grou s of r-;,g;ed to recogn'ze:j",;tgisto;;$- 8 relaxation t This ap roach is robust because it bases its decisions on a E 1 the available information, and the location strategy is straightforward because all the features are used.
BOOTSTRAP STEREO Marsha Jo Hannah Lockheed Palo Alto Research Laboratory Department 52-53, Building 204 3251 Hanover Street, Palo Alto, CA 94304 ABSTRACT Lockheed has been working on techniques for navigation of an autonomous aerial vehicle using passively sensed images. One technique which shows promise is bootstrap stereo, in which the vehicle's position is determined from the perceived locations of known ground control points, then two known vehicle camera positions are used to locate corresponding image points on the ground, creating new control points. This paper describes the components of bootstrap stereo. I INTRODUCTION Before the advent of sophisticated navigation aids such as radio beacons, barnstorming pilots relied primarily on visual navigation. A pilot would lookart the window of his airplane, see landmarks below him, and know where he was.
ABSTRACT We deal with the inference of environmental information (position and velocity) from a sequence of images formed during relative motion of an observer and the environment. A simple method is used to transform relations between environmental points into equations expressed in terms of constants determined from the images and unknown depth values. This is used to develop equations for environmental inference from several cases of rigid body motion, some having direct solutions. Also considered are the problems of non-unique solutions and the necessity of decomposing the inferred motion into natural components. Inference from optic flow is based upon the analysis of the relative motions of points in images formed over time.
Sticks, Plates, and Blobs: A Three-Dimensional Object Representation for Scene Analysis Linda G. Shapiro Prasanna G. Mulgaonkar John D. Moriarty Robert M. Haralick Virginia Polytechnic Institute and State University Department of Computer Science how the parts fit together. Our models have three ABSTRACT kinds of three-dimensional Parts: sticks, Plates. In this paper, we describe a relational modeling technique which categorizes three-dimensional objects at a gross level. These models may then be used to classify and recognize two dimensional views of the object, in a scene analysis system. I. Introduction The recognition of three-dimensional objects from two-dimensional views is an important and still largely unsolved problem in scene analysis.