Showing papers presented at "British Machine Vision Conference in 1990"

RAPID - a video rate object tracker.

Abstract: Three-dimensional (3D) Model-Based Vision enables observed image features to be used to determine the pose (ie. position and attitude) of a known 3D object with respect to the camera (or alternatively, the viewpoint of the camera with respect to the model) [1]. The knowledge concerning the object that is used to perform Model-Based Vision is the 3D locations of salient and easily observed object features, such as corners and high-contrast edges (eg. surface marking sand crease edges).

Feature grouping by 'relocalisation' of eigenvectors of the proximity matrix.

Abstract: We describe a widely applicable method of grouping or clustering image features (such as points, lines, corners, flow vectors and the like). It takes as input a "proximity matrix" H a square, symmetric matrix of dimension N (where N is the number of features). The element i,j of H is an initial estimate of the "proximity" between the ith and yth features. As output it delivers another square symmetric matrix S whose i-)th element is near to, or much less than unity according as features i and j are to be assigned to the same or different clusters.

Symbolic Image Matching by Simulated Annealing

Abstract: In this paper we suggest an optimization approach to visual matching. We assume that the information available in an image may be conveniently represented symbolically in a relational graph. We concentrate on the problem of matching two such graphs. First we derive a cost function associated with graph matching and more precisely associated with relational subgraph isomorphism and with maximum relational subgraph matching. This cost function is well suited for optimization methods such as simulated annealing. We show how the graph matching problem is easily cast into a simulated annealing algorithm. Finally we show some preliminary experimental results and discuss the utility of this graph matching method in computer vision in general.

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.

Determining motion from 3D line segment matches: a comparative study.

Abstract: Motion estimation is a very important problem in dynamic scene analysis. Although it is easier to estimate motion parameters from 3D data than from 2D images, it is not trivial, since the 3D data we have are almost always corrupted by noise. A comparative study on motion estimation from 3D line segments is presented. Two representations of line segments and two representations of rotation are described. With different representations of line segments and rotation a number of methods for motion estimation are presented, including the extended Kalman filter a general minimization process and the singular value decomposition. These methods are compared using both synthetic and real data obtained by a trinocular stereo. It is observed that the extended Kalman filter with the rotation axis representation of rotation is preferable. Note that all methods discussed can be directly applied to 3D point data.

Robust estimation of shape parameters

Abstract: We investigate the use of Robust Estimation in an application requiring the accurate location of the centres of circular objects in an image. A common approach used throughout computer vision for extracting shape information from a data set is to fit a feature model using the Least Squares method. The well known sensitivity of this method to outliers is traditionally accommodated by outlier rejection methods. These usually consist of heuristic applications of model templates or data trimming. Robust Estimation offers a theoretical framework for assessing such rejection schemes, and more importantly, provides an approach to parameter estimation in contaminated data distributions capable of greater accuracy.

Optimal combination of stereo camera calibration from arbitrary stereo images.

Abstract: Many stereo correspondence algorithms require relative camera geometry, as the epipolar constraint is fundamental to their matching processes. We intend to build a eye/head camera rig to mount on the mobile platform COMODE to enhance the abilities of the TINA system to recover 3D geometry from its environment. Thus we need to be able to associate camera geometry with particular head configurations. Generic calibration of such a system would require the ability to compute camera geometry from arbitrary stereo images. A system which solves this problem using an established corner detector combined with a robust stereo matching algorithm and a variational solution for the camera geometry is described.

A probabilistic approach to the Hough Transform.

Abstract: The mathematical principles come from the "Maximum Likelihood Method" [2, 3], used in probability theory for the determination of distribution parameters from experimental data. The Maximum Likelihood analysis leads to the definition of the Probabilistic Hough Transform, which is a likelihood function. If certain assumptions are made about the error characteristics, the PHT is very close to conventional Hough Transforms. If, in a particular application, these assumptions are a reasonable approximation, good results are usually obtained using standard Hough methods. However, where these assumptions are far from the truth, the Hough Transform will not work well, and steps should be taken to improve the model of input feature errors, such as filtering the Hough space, or incrementing an extended region instead of just the voting space. As a last resort, the full PHT can be computed, but this is much more computationally expensive than conventional Hough methods.

An Uncalibrated Stereo Visual Servo System

Abstract: We describe a robotic system composed of a manipulator and two cameras. We use the vision system to guide the robot hand to a visible target. The camera positions are known only approximately. Our system does not use the details of the kinematics of the manipulator. There is no common frame of reference linking vision system, workspace and robot hand. The stereo vision system gives information in terms of picture coordinates. This information is used to control a three degree of freedom robot manipulator in a straightforward and robust fashion, in terms of lines and points visible in the images. We describe the implementation with which we tested this idea.

An Application of Active Contour Models to Head Boundary Location

Abstract: Locating the boundary of a head in a head-and-shoulders image is an important problem in model based coding. An approach to this problem using adaptive contour models or 'snakes' is presented. The paper provides a tutorial introduction to the theory of snakes and shows in some detail how they may be implemented using a finite difference method. Some experimental results are presented showing snakes locating the head in a set of test images.

Integrating stereo and photometric stereo to monitor the development of glaucoma.

Abstract: Progress on a system to monitor the development of glaucoma by measuring the topography of the optic disk is reported. The need for an accurate method for doing this using passive vision is explained. Sparse depth measurements from stereo matching of blood vessels provide insufficient constraint for reconstructing the surface of the optic disk. Shape from shading has to contend with a spatially-varying albedo. We show how a particular arrangement of fundus cameras allows us to apply a technique akin to homomorphic filtering to recover estimates of δz δx that can be smoothed by appropriate regularization. Stereo is integrated with these photometric stereo depth estimates. Examples of artificial object and optic disk surface reconstructions are presented.

Construction of a Scale-Space Primal Sketch

Abstract: We present a multi-scale representation of grey-level shape, called scale-space primal sketch, that makes explicit features in scale-space as well as the relations between features at different levels of scale. The representation gives a qualitative description of the image structure that allows for extraction of significant image structure --- stable scales and regions of interest --- in a solely bottom-up data-driven manner. Hence, it can be seen as preceding further processing, which can then be properly tuned. Experiments on real imagery demonstrate that the proposed theory gives perceptually intuitive results.

Combining cues for mammographic abnormalities

Abstract: suspected abnormalities. Previous attempts to automate the detection of microcalcifications have used sequences of progressively more sophisticated methods to refine a set of candidates e.g. [2], though clinically acceptable error rates have not yet been achieved. Screening for breast cancer involves searching for subtle abnormalities in a large number of complex images, a task for which the specificity of human interpreters is known to be poor. We are aiming to improve screening performance by providing radiologists with machine assistance in the detection of clinically significant features. The first stage of the detection process is the generation of a set of cues to indicate potential abnormalities. We generally select a cue method for a particular task because it responds to a known property of the target. However, cue generators also respond to non-targets which share that target property. By combining evidence from a range of cues associated with different target properties we can increase the specificity of detection in noisy or cluttered images. We have performed experiments which demonstrate this. Two cue generators were applied to a set of 20 digitised image patches. Onand off-target distributions were collected for each image and accumulated across the data set on a leave-one-out basis. Each cue image was then transformed into a log-likelihood image, enabling evidence from the different cue generators to be combined simply by image addition. Results of an evaluation of single and combined cue methods are presented.

Active Contours in Medical Image Processing using a Networked SIMD Array Processor

Abstract: This paper reports work carried out within the BIOLAB consortium of the CEC Advanced Informatics in Medicine (AIM) programme. The overall purpose of the work is to assess the potential for commercially available parallel computer systems to enhance practical applications of medical image analysis. Here, we report methods developed for using the Active Memory Technology (AMT) DAP as a "compute server" to a network of workstations. We report three technical developments.

Transformational invariance - a primer.

Abstract: The shape of objects seen in images depends on the viewpoint. This effect confounds recognition. We demonstrate a theoretical framework within which it is possible to construct descriptors for curves which do not vary with viewpoint. These descriptors are known as invariants. We use this framework to construct invariant shape descriptors for plane curves. These invariant shape descriptors make it possible to recognise plane curves, without explicitly determining the relationship between the curve reference frame and the camera coordinate system, and can be used to index quickly and efficiently into a large model base of curves. Many of these ideas are demonstrated by experiments on real image data.

Active intelligent vision using the dynamic generalized Hough Transform.

Abstract: Parametric transformation is a powerful tool in shape analysis which gives good results even in the presence of noise and occlusion. Major shortcomings of the technique are excessive storage requirements, computational complexity and the need to initiate a separate transformation process with respect to each feature under detection. In addition, standard parametric transformation processes the entire image treating each image point independently. The proposed method selectively segments the image and provides a feedback mechanism linking image and transform space. Decisions are made concerning the probable instance of shape under detection, the viability of processing and the need to gather further evidence from the image. The processing monitors itself and adapts to focus on areas of interest avoiding computationally expensive global processing. Low level processing and high level decision making are provided with a window of communication not previously available using such parametric transform methods. The method is applied and illustrated with respect to the detection of circles and elipses.

Kalman filtering of pose estimates in applications of the RAPID video rate tracker.

Abstract: RAPID is a video-rate model based tracker which generates refined estimates of an object's position and orientation (pose) given approximate initial estimates. A practical application of this technique requires (1) prediction from-frame-to-frame of the pose of the object being tracked, to accommodate realistic target movement and (2) temporal integration of pose estimates to reduce measurement noise. These needs are both satisfied by a Kalman filter. In order to apply a Kalman filter, however, we first construct statistical models of both the apparent motion of the object between frames and also of the accuracy of pose measurements made at each processed frame. The filtered tracker output provides a robust estimate of object pose at video rate when implemented in software running on a standard mini-computer. The capabilities of this technique are demonstrated by application to the task of monitoring the pose of an unmanned aircraft during its approach to an airfield and during landing.

Edge-region integration for segmentation of MR images.

Abstract: A data-driven segmentation scheme is described that integrates edges and smooth regions. Edges are detected first and then used to guide the action of a dualresolution agglomeratine •parametric region detector. The scheme is applied to a variety of MR images. A comparison is made of segmentations produced by the scheme and by medical experts. The results are better than those obtained from the edge detection process alone.

3-D cues from a single view: detection of elliptical arcs and model-based perspective backprojection.

Abstract: complexity of the scene which can be formed by several partially occluded objects (e.g. the bin-picking task). We describe the use of geometric reasoning and proI n t h e l i t e r a t u r * J t i s P° s s i b l e t o find <&* a TMmbe o f jective geometry to infer 3-D information about cirm e t h o d s f o r e l h P t i c a l 8 b u t t h e P r o b l e m o f d e a l i n 8 cles which project elliptical curves onto one 2-D imefficiently with real images has been often ignored. age. A robust and efficient method to extract elliptical arcs starting from chains of line segments obtained by the polygonal approximation of edge curves is presented. Then an algorithm that solves analytically the problem of the perspective inversion of an elliptical arc from its projection onto the image plane is outlined. Finally experimental results on real images of mechanical parts are reported; therefore, current drawbacks and future extensions of this work are discussed. Preprocessing f Polygonal (Approximation \ ^ Chains

A Knowledge Based System for Measuring Faces

Abstract: Recognising faces is now an attractive computer vision task. Methods used for machine parts; recognizing straight lines or circular arcs, and the use of detailed geometric models, are not directly relevant, yet with many human faces to recognize, it is worthwhile building a system solely for this task. The system described here uses a blackboard architecture and aims to obtain a set of measurements from an image of a face. A key feature is the use of knowledge about faces, including a statistical knowledge base derived from detailed measurements of a large number of sample images.

Shape from texture: textural invariance and the problem of scale in perspective images of textured surfaces.

Abstract: If texels on a given surface can be identified in an image then the orientation of that surface can be obtained [1]. Unfortunately there is no known procedure for identifying texels for arbitrary textures. Even if the size and shape of texels on the surface is invariant with regard to position, perspective projection ensures that the size and shape of the corresponding image texels will vary by orders of magnitude.

A head called Richard

Abstract: This paper describes two preliminary experiments concerned with the construction of a robot head. The initial design and research is aimed at producing a system with two cameras and two microphones on a system capable of operating with the same degrees of freedom and reflex times as its biological counterpart. Whilst the primary goal of the project is develop an anthropomorphic system with the sensory reflex capabilities of a human head, the system will also contain some non-anthropomorphic components. The most obvious of the non-anthropomorphic components is a spatially and temporally programmable light source. Some preliminary results are presented. use this information to direct autonomous behaviour (Brooks, 1989) but these reflexes have not been coupled with vision systems for integrated tracking or gaze control.

A fast algorithm for computing optic flow and its implementation on a Transputer array.

Abstract: to compute the optic flow along intensity change curves. The flow is computed locally at corners, determined as edge points of high curvature, by cross correlation over time. The (full) flow is then computed along edge contours by means of a combined wave/diffusion process. MCMAP has been implemented on a Transputer array using a mixed paradigm of a pipeline and a processor farm. A fast algorithm for computing optic flow and its implementation on a Transputer array

Epipolar geometry for trinocular active range-sensors.

Abstract: of these for certain types of scene, so we hope to be able to label lines and nodes successfully without any type of coding. We investigate recent ideas in the design of irinocular active range-sensors. Such devices have the advantage of freedom from mechanical scanning, and rapid image capture. The main technical problem is overcoming the correspondence problem. This requires careful geometric design to take account of epipolar geometry and thorough modelling of image-measurement error. We present a novel design that, so far, seems to work well. Curiously it involves setting up the projectorcamera geometry to be degenerate — so that depth computation is ill-conditioned — and then backing off a little.

Digital or analog Hough Transform

Abstract: A variation of the Hough Transform that is aimed at detecting digital lines has been recently suggested. Other Hough algorithms are intended to detect straight lines in the analog pre-image. These approaches are analyzed and compared in terms of the relation between the achievable resolution and the required number of accumulators, using a definition of resolution that is based on the Geometric Probability measure of straight lines. It is shown that the ‘analog’ approach is greatly superior in high resolution applications, where a ‘digital’ Hough Transform would generally require an infeasibly large number of accumulators.

Optimal probabilistic relaxation labeling.

Abstract: This paper investigates the theoretical limits of probabilistic relaxation labeling (PRL), applied to per-pixel contextual image classification. The performance of a scheme which is defined to be optimal (within a class of PRL schemes) is studied, and found to fall short of that theoretically obtainable by directly considering all the original a posteriori probabilities (PPs) in the image. Lt is also found that an optimal scheme must use different updating functions at each iteration, and that these functions will depend on the distributions of the original per-pixel data.

Shape verification using belief updating.

Abstract: This paper addresses the problem of2D shape representation and its application to object verification. We show how knowledge of shape can be integrated in a principled manner with low-level evidence such as an estimate of object position and an edge strength map. We begin by considering the role of shape in image interpretation and discuss the criteria which should be applied in assessing representations of shape. We propose new criteria, particularly as regards the ability to model variability, and describe a Chord Length Distribution (CLD) representation of shape which possesses many desirable properties. We show how the CLD representation can be used in an iterative belief-updating scheme for object location and verification. We give experimental results which demonstrate the feasibility of the method and discuss future developments.

Using a combined stereo/temporal matcher to determine ego-motion.

Abstract: determine the ego-motion and for this we combine the transformation estimation suggested by Faugeras [1] with the data weighting method suggested by Kiang [4]. Once the camera motion is determined it is possible to combine positional information of each feature to increase their localisation accuracy. A system has been developed for integrating information from sequences of stereo images suitable for use in visual control. The method exploits multiple sources of information to obtain a subset of correctly matched corner features in temporal pairs of stereo images. These have been used to determine the egomotion of a stereo camera system and to improve position estimates of these features. The algorithm is demonstrated on images of a real scene which would be expected to present stereo or temporal matching algorithms with matching difficulties.

A Parallel Path Planning Algorithm for Mobile Robots

Abstract: This paper presents a path planning algorithm for mobile robots. We introduce a parallel search approach which is based on a regular grid representation of the map. The search is formulated as a cellular automaton by which local inter-cell communication rules are defined. The algorithm is made adaptive by utilising a multiresolution representation of the map. It is implemented on AMT DAP 510, which is a SIMD machine.