Stephen Pollard

PMF: A Stereo Correspondence Algorithm Using a Disparity Gradient Limit

The advantages of solving the stereo correspondence problem by imposing a limit on the magnitude of allowable disparity gradients are examined. It is shown how the imposition of such a limit can provide a suitable balance between the twin requirements of disambiguating power and the ability to deal with a wide range of surfaces. Next, the design of a very simple stereo algorithm called PMF is described. In conjunction with certain other constraints used in many other stereo algorithms, PMF employs a limit on allowable disparity gradients of 1, a value that coincides with that reported for human stereoscopic vision. The excellent performance of PMF is illustrated on a series of natural and artificial stereograms. Finally, the differences between the theoretical justification for the use of disparity gradients for solving the stereo correspondence problems presented in the paper and others that exist in the stereo algorithm literature are discussed.

Curve matching and stereo calibration

Abstract The topological obstacles to the matching of smooth curves in stereo images are shown to occur at epipolar tangencies. Matching is possible when these tangencies satisfy certain projective constraints (the tangent lines form corresponding pencils) and metric contraints dependent on the camera geometry. Such points are good matching primitives, even when the image curves correspond to smooth surface profiles. An iterative scheme for improving camera calibration based on these results is derived, and performance demonstrated on real data.

Matching geometrical descriptions in three-space

Abstract A matching strategy for combining two or more three-space descriptions, obtained here from edge-based binocular stereo, of a scene is discussed. The scheme combines features of a number of recent model-matching algorithms with heuristics aimed to reduce the space of potential rigid transformations that relate scene descriptions.

Geometrical modeling from multiple stereo views

The components of the Sheffield Artificial Intelligence Vision Research Unit (AIVRU) three-dimensional (3D) vision sys tem, which currently supports model-based object recognition and location, are described. Its potential for robotics applica tions is demonstrated by its guidance of a Universal Machine Intelligence robot arm in a pick-and-place task. The system comprises (1) the recovery of a sparse depth map using edge- based, passive stereo triangulation; (2) the grouping, descrip tion, and segmentation of edge segments to recover a 3D representation of the scene geometry in terms of straight lines and circular arcs; (3) the statistical combination of 3D de scriptions for object model creation from multiple stereo views and the propagation of constraints for within-view re finement ; and (4) the matching of 3D wireframe object models to 3D scene descriptions in order to recover an initial estimate of their position and orientation.

TINA: a 3D vision system for pick and place

Abstract The paper describes the Sheffield AIVRU 3D vision system for robotics. The system currently supports model-based object recognition and location; its potential for robotics applications is demonstrated by its guidance of a UMI robot arm in a pick-and-place task. The system comprises: recovery of a sparse depth map using edgebased passive stereo triangulation; grouping, description and segmentation of edge segments to recover a 3D description of the scene geometry in terms of straight lines and circular arcs; statistical combination of 3D descriptions for the purpose of object model creation from multiple stereo views, and the propagation of constraints for within-view refinement; and matching 3D wireframe models to 3D scene descriptions to recover an initial estimate of their position and orientation.

Optimal combination of multiple sensors including stereo vision

Abstract The statistical combination of information from multiple sources is considered. The particular needs of the target application, stereo vision, require that the formulation be adequate to deal with highly correlated errors and constraints, and that it deal naturally with geometrical data.

Multiprocessor 3D vision system for pick and place

Abstract A 3D vision system implemented upon a locally developed transputer-based hybrid parallel processing engine named MARVIN (Multiprocessor ARchitecture for VIsioN), hosted by a SUN workstation, is described. In addition to the recovery of scene descriptions from edge-based binocular stereo, the system incorporates a parallel model matching algorithm which is able to accurately locate modelled objects within such scenes. The competence of this vision system is demonstrated by visually guiding a robot arm to pick up various objects in a cluttered scene with a total processing time of approximately ten seconds.

Recovering partial 3D wire frames descriptions from stereo data

Abstract The design of modules in the current version of the TINA stereo based 3D vision system responsible for the recovery of geometric descriptions and their subsequent integration into partial wire frame models are described. The approach differs considerably from that described in an earlier version of the TINA system 1,2 . The strategy for the construction of partial wire frame descriptions is broken into the following three processing stages: 1. 1 Stereo processing: this is edged-based, and uses camera calibration information to provide a restriction on allowable matches between edges extracted from the left and right images. The algorithm does not, however, at this stage recover an accurate estimate of disparity of each pair of matched edges. 2. 2 Geometrical description: this relies on the identification of higher level 2D primitives from the edge strings in one, other or both images and their subsequent combination with stereo matching, and calibration data to allow the recovery of 3D primitives. 3. 3 Wire frame generation: based upon 2D and 3D proximity and connectivity heuristics potential wire frame components are identified.

MARVIN and TINA: a multiprocessor 3-D vision system

We describe a working multi-transputer stereo vision system which exploits various forms of parallelism in a number of visual competences upon a specialized hardware architecture that provides distributed video datapaths to the processor array. We exploit stereo and spatial parallelism to recover 3-D scene descriptions from passive stereo vision. To recognize and recover the positions of modelled objects, we exploit featural parallelism at both the object and sub-object level. We also describe a real-time object-tracking algorithm that exploits featural parallelism in the concurrent tracking of a set of object features. These competences have been integrated into a general-purpose architecture and we present some performance results and descriptions of the use of this vision engine in two application domains.

Robust Recovery of 3D Ellipse Data

This paper is concerned with robust, accurate and computationally tractable methods for the automatic recovery of 3D ellipse data from edge based stereo. The processing paradigm relies heavily on the 2D image as a rich and robust source of scene feature hypotheses (in this case ellipses). Rather than attempt to recover 3D scene descriptions by grouping unstructured estimates of disparity and/or depth, a processes of automatic 2D feature hypothesis is used in conjunction with an appropriate disparity grouping constraint (in the case of ellipse hypotheses we use an affine disparity plane constraint) to recover more accurate 3D scene descriptors.