Daniela Hall

Head Pose estimation on low resolution images

This paper addresses the problem of estimating head pose over a wide range of angles from low-resolution images. Faces are detected using chrominance-based features. Grey-level normalized face imagettes serve as input for linear auto-associative memory. One memory is computed for each pose using a Widrow-Hoff learning rule. Head pose is classified with a winner-takes-all process. We compare results from our method with abilities of human subjects to estimate head pose from the same data set. Our method achieves similar results in estimating orientation in tilt (head nodding) angle, and higher precision for estimating orientation in the pan (side-to-side) angle.

MagicBoard: A contribution to an intelligent office environment

Abstract In this paper, we describe an augmented reality tool for collaborative work called the MagicBoard. The MagicBoard is based on an ordinary white board which has been enhanced by a video-projector and a steerable camera. A supervisor coordinates the cooperation of several modules including gesture recognition, finger tracking and white board scanning for digitalisation of the content. The gesture recognition module uses an approach based on local spatio-temporal appearance of activities. The tracking module is designed for use with cluttered backgrounds and variable lighting conditions. The white board scanner eliminates global luminosity differences by adaptive thresholding and the result can serve to digitise the content of the board. The supervisor is based on a rule-based architecture and is easily extendable. The selected modules fit together to a compact system, that largely increases the functionality of a white board and makes it a useful tool in the future office environments.

Facial features detection robust to pose, illumination and identity

This paper addresses the problem of automatic detection of salient facial features. Face images are described using local normalized Gaussian receptive fields. Face features are learned using a clustering of the Gaussian derivative responses. We have found that a single cluster provides a robust detector for salient facial features robust to pose, illumination and identity. In this paper we describe how this cluster is learned and which facial features have found to be salient

Object Recognition Using Coloured Receptive Fields

This paper describes an extension of a technique for the recognition and tracking of every day objects in cluttered scenes. The goal is to build a system in which ordinary desktop objects serve as physical icons in a vision based system for man-machine interaction. In such a system, the manipulation of objects replaces user commands. A view-variant recognition technique, developed by the second author, has been adapted by the first author for a problem of recognising and tracking objects on a cluttered background in the presence of occlusions. This method is based on sampling a local appearance function at discrete viewpoints by projecting it onto a vector of receptive fields which have been normalised to local scale and orientation. This paper reports on the experimental validation of the approach, and of its extension to the use of receptive fields based on colour. The experimental results indicate that the second authors technique does indeed provide a method for building a fast and robust recognition technique. Furthermore, the extension to coloured receptive fields provides a greater degree of local discrimination and an enhanced robustness to variable background conditions. The approach is suitable for the recognition of general objects as physical icons in an augmented reality.

Local Scale Selection for Gaussian Based Description Techniques

This paper addresses the problem of the local scale parameter selection for recognition techniques based on Gaussian derivatives. Patterns are described in a feature space of which each dimension is a scale and orientation normalized receptive field (a unit composed of normalized Gaussian-based filters). Scale invariance is obtained by automatic selection of an appropriate local scale [Lin98b] and followed by normalisation of the receptive field to the appropriate scale. Orientation invariance is obtained by the determination of the dominant local orientation and by steering the receptive fields to this orientation. Data is represented structurally in a feature space that is designed for the recognition of static object configurations. In this space an image is modeled by the vectorial representation of the receptive field responses at each pixel, forming a surface in the feature space. Recognition is achieved by measuring the distance between the vector of normalized receptive fields responses of an observed neighborhood and the surface point of the image model. The power of a scale equivariant feature space is validated by experimental results for point correspondences in images of different scales and the recognition of objects under different view points.

Brand identification using Gaussian derivative histograms

Abstract.In this article, we describe a module for the identification of brand logos from video data. A model for the visual appearance of each logo is generated from a small number of sample images using multidimensional histograms of scale-normalized chromatic Gaussian receptive fields. We compare several identification techniques based on multidimensional histograms. Each of the methods displays high recognition rates and can be used for logo identification. Our method for calculating scale-normalized Gaussian receptive fields has linear computational complexity and is thus well adapted to a real-time system. However, with the current generation of microprocessors we obtain at best only two images per second when processing a full PAL video stream. To accelerate the process, we propose an architecture that combines fast detection, reliable identification, and fast tracking for speedup. The resulting real-time system is evaluated using video streams from sports Formula 1 races and football.

Automatic parameter regulation of perceptual systems

Abstract Changes in environmental conditions frequently degrade the performance of perceptual systems. This article proposes a system architecture with a control component that auto-regulates parameters to provide a reduction in the sensitivity to environmental changes. We demonstrate the benefit of this architecture using the example of a long-term tracking system. The control component consists of modules for auto-critical evaluation, for auto-regulation of parameters and for error recovery. Both modules require a measure of the goodness of system output with respect to a scene reference model. We describe the generation of the scene reference model and propose measures for the model quality and for the goodness of system output in form of measurement trajectories. Our self-adaptive tracking system achieves better recall than a manually tuned tracking system on a public benchmark data set.

Statistical Gesture Recognition Through Modelling of Parameter Trajectories

The recognition of human gestures is a challenging problem that can contribute to a natural man-machine interface. In this paper, we present a new technique for gesture recognition. Gestures are modelled as temporal trajectories of parameters. Local sub-sequences of these trajectories are extracted and used to define an orthogonal space using principal component analysis. In this space the probabilistic density function of the training trajectories is represented by a multidimensional histogram, which builds the basis for the recognition. Experiments on three different recognition problems show the general utility of the approach.

Comparative study of people detection in surveillance scenes

We address the problem of determining if a given image region contains people or not, when environmental conditions such as viewpoint, illumination and distance of people from the camera are changing. We develop three generic approaches to discriminate between visual classes: ridge-based structural models, ridge-normalized gradient histograms, and linear auto-associative memories. We then compare the performance of these approaches on the problem of people detection for 26 video sequences taken from the CAVIAR database.

Automatic parameter regulation for a tracking system with an auto-critical function

In this article we propose an architecture of a tracking system that can judge its own performance by an auto-critical function. Performance drops can be detected which trigger an automatic parameter regulation module. This regulation module is an expert system that searches a parameter setting with better performance and returns it to the tracking system. With such an architecture, a robust tracking system can be implemented which automatically adapts its parameters in case of changes in the environmental conditions. This article opens a way to self-adaptive systems in detection and recognition.