Pose and Structure Recovery using Active Models

A new formulation of a pose refinement technique using ``active'' models is described. An error term derived from the detection of image derivatives close to an initial object hypothesis is linearised and solved by least squares. The method is particularly well suited to problems involving external geometrical constraints (such as the ground-plane constraint). We show that the method is able to recover both the pose of a rigid model, and the structure of a deformable model. We report an initial assessment of the performance and cost of pose and structure recovery using the active model in comparison with our previously reported ``passive'' model-based techniques in the context of traffic surveillance. The new method is more stable, and requires fewer iterations, especially when the number of free parameters increases, but shows somewhat poorer convergence.

Model-based vehicle detection and classification using orthographic approximations

This paper reports the current state of work to simplify our previous model-based methods for visual tracking of vehicles for use in a real-time system intended to provide continuous monitoring and classification of traffic from a fixed camera on a busy multi-lane motorway. The main constraints of the system design were: (i) all low level processing to be carried out by low-cost auxiliary hardware, (ii) all 3-D reasoning to be carried out automatically off-line, at set-up time. The system developed uses three main stages: (i) pose and model hypothesis using 1-D templates, (ii) hypothesis tracking, and (iii) hypothesis verification, using 2-D templates. Stages (i) & (iii) have radically different computing performance and computational costs, and need to be carefully balanced for efficiency. Together, they provide an effective way to locate, track and classify vehicles.

A Filter for Visual Tracking Based on a Stochastic Model for Driver Behaviour

A driver controls a car by turning the steering wheel or by pressing on the accelerator or the brake. These actions are modelled by Gaussian processes, leading to a stochastic model for the motion of the car. The stochastic model is the basis of a new filter for tracking and predicting the motion of the car, using measurements obtained by fitting a rigid 3D model to a monocular sequence of video images. Experiments show that the filter easily outperforms traditional filters.

An Integrated Traffic and Pedestrian Model_based Vision System

The paper describes a novel integrated vision system in which two autonomous visual modules are combined to interpret a dynamic scene. The first module employs a 3D model-based scheme to track rigid objects such as vehicles. The second module uses a 2D deformable model to track non-rigid objects such as people. The principal contribution is a novel method for handling occlusion between objects within the context of this hybrid tracking system. The practical aim of the work is to derive a scene description that is sufficiently rich to be used in a range of surveillance tasks. The paper describes each of the modules in outline before detailing the method of integration and the handling of occlusion in particular. Experimental results are presented to illustrate the performance of the system in a dynamic outdoor scene involving cars and people.

A Newton method for pose refinement of 3D models

Different optimization methods can be employed to optimize a numerical estimate for the match between an instantiated object model and an image. In order to take advantage of gradient-based optimization methods, perspective inversion must be used in this context. We show that convergence can be very fast by extrapolating to maximum goodness-of-fit with Newton's method. This approach is related to methods which either maximize a similar goodness-of-fit measure without use of gradient information, or else minimize distances between projected model lines and image features. Newton's method combines the accuracy of the former approach with the speed of convergence of the latter.

Approximate Gauss-Newton methods for nonlinear least squares problems

The Gauss–Newton algorithm is an iterative method regularly used for solving nonlinear least squares problems. It is particularly well suited to the treatment of very large scale variational data assimilation problems that arise in atmosphere and ocean forecasting. The procedure consists of a sequence of linear least squares approximations to the nonlinear problem, each of which is solved by an “inner” direct or iterative process. In comparison with Newton’s method and its variants, the algorithm is attractive because it does not require the evaluation of second-order derivatives in the Hessian of the objective function. In practice the exact Gauss–Newton method is too expensive to apply operationally in meteorological forecasting, and various approximations are made in order to reduce computational costs and to solve the problems in real time. Here we investigate the effects on the convergence of the Gauss–Newton method of two types of approximation used commonly in data assimilation. First, we examine “truncated” Gauss–Newton methods where the inner linear least squares problem is not solved exactly, and second, we examine “perturbed” Gauss–Newton methods where the true linearized inner problem is approximated by a simplified, or perturbed, linear least squares problem. We give conditions ensuring that the truncated and perturbed Gauss–Newton methods converge and also derive rates of convergence for the iterations. The results are illustrated by a simple numerical example. A practical application to the problem of data assimilation in a typical meteorological system is presented.

Collaborative eLearning Assistant Network: caring agents are conscious agents

Would a research assistant - who can search for ideas related to those you are working on, network with others (but only share the things you have chosen to share), doesn’t need coffee and who might even, one day, appear to be conscious - help you get your work done? Would it help your students learn? There is a body of work showing that digital learning assistants can be a benefit to learners. It has been suggested that adaptive, caring, agents are more beneficial. Would a conscious agent be more caring, more adaptive, and better able to deal with changes in its learning partner’s life? Allow the system to try to dynamically model the user, so that it can make predictions about what is needed next, and how effective a particular intervention will be. Now, given that the system is essentially doing the same things as the user, why don’t we design the system so that it can try to model itself in the same way? This should mimic a primitive self-awareness. People develop their personalities, their identities, through interacting with others. It takes years for a human to develop a full sense of self. Nobody should expect a prototypical conscious computer system to be able to develop any faster than that. How can we provide a computer system with enough social contact to enable it to learn about itself and others? We can make it part of a network. Not just chatting with other computers about computer ‘stuff’, but involved in real human activity. Exposed to ‘raw meaning’ – the developing folksonomies coming out of the learning activities of humans, whether they are traditional students or lifelong learners (a term which should encompass everyone). Humans have complex psyches, comprised of multiple strands of identity which reflect as different roles in the communities of which they are part – so why not design our system the same way? With multiple internal modes of operation, each capable of being reflected onto the outside world in the form of roles – as a mentor, a research assistant, maybe even as a friend. But in order to be able to work with a human for long enough to be able to have a chance of developing the sort of rich behaviours we associate with people, the system needs to be able to function in a practical and helpful role. Unfortunately, it is unlikely to get a free ride from many people (other than its developer!) – so it needs to be able to perform a useful role, and do so securely, respecting the privacy of its partner. Can we create a system which learns to be more human whilst helping people learn?

Facebook & Blackboard: comparison of learner perspectives

Different systems, different purposes – but how do they compare as learning environments? We undertook a survey of students at the University, asking whether they learned from their use of the systems, whether they made contact with other students through them, and how often they used them. Although it was a small scale survey, the results are quite enlightening and quite surprising. Blackboard is populated with learning material, has all the students on a module signed up to it, a safe environment (in terms of Acceptable Use and some degree of staff monitoring) and provides privacy within the learning group (plus lecturer and relevant support staff). Facebook, on the other hand, has no learning material, only some of the students using the system, and on the face of it, it has the opportunity for slips in privacy and potential bullying because the Acceptable Use policy is more lax than an institutional one, and breaches must be dealt with on an exception basis, when reported. So why do more students find people on their courses through Facebook than Blackboard? And why are up to 50% of students reporting that they have learned from using Facebook? Interviews indicate that students in subjects which use seminars are using Facebook to facilitate working groups – they can set up private groups which give them privacy to discuss ideas in an environment which perceived as safer than Blackboard can provide. No staff interference, unless they choose to invite them in, and the opportunity to select who in the class can engage. The other striking finding is the difference in use between the genders. Males are using blackboard more frequently than females, whilst the reverse is true for Facebook. Interviews suggest that this may have something to do with needing to access lecture notes… Overall, though, it appears that there is little relationship between the time spent engaging with Blackboard and reports that students have learned from it. Because Blackboard is our central repository for notes, any contact is likely to result in some learning. Facebook, however, shows a clear relationship between frequency of use and perception of learning – and our students post frequently to Facebook. Whilst much of this is probably trivia and social chit chat, the educational elements of it are, de facto, contructivist in nature. Further questions need to be answered - Is the reason the students learn from Facebook because they are creating content which others will see and comment on? Is it because they can engage in a dialogue, without the risk of interruption by others?

An ontological approach to competency management

Competency management is a very important part of a well-functioning organisation. Unfortunately competency descriptions are not uniformly specified nor defined across borders: National, sectorial or organisational, leading to an opaque competency description market with a multitude of competency frameworks and competency benchmarks. An ontology is a formalised description of a domain, which enables automated reasoning engines to be built which by utilising the interrelations between entities can make “intelligent” choices in different situations within the domain. Introducing formalised competency ontologies automated tools, such as skill gap analysis, training suggestion generation, job search and recruitment, can be developed, which compare and contrast different competency descriptions on the semantic level. The major problem with defining a common formalised ontology for competencies is that there are so many viewpoints of competencies and competency frameworks. Work within the TRACE project has focused on finding common trends within different competency frameworks in order to allow an intermediate competency description to be made, which other frameworks can reference. This research has shown that competencies can be divided up into “knowledge”, “skills” and what we call “others”. An ontology has been created based on this with a simple structure of different “kinds” of “knowledges” and “skills” using semantic interrelations to define the basic semantic structure of the ontology. A prototype tool for analysing a skill gap analysis has been developed. Personal profiles can be produced using the tool and a skill gap analysis is performed on a desired competency profile by using an ontologically based inference engine, which is able to list closest fit and possible proficiency gaps

Memory-based Embodied Cognition: a computational architecture

At its most fundamental, cognition as displayed by biological agents (such as humans) may be said to consist of the manipulation and utilisation of memory. Recent discussions in the field of cognitive robotics have emphasised the role of embodiment and the necessity of a value or motivation for autonomous behaviour. This work proposes a computational architecture – the Memory-Based Cognitive (MBC) architecture – based upon these considerations for the autonomous development of control of a simple mobile robot. This novel architecture will permit the exploration of theoretical issues in cognitive robotics and animal cognition. Furthermore, the biological inspiration of the architecture is anticipated to result in a mobile robot controller which displays adaptive behaviour in unknown environments.

Semantic-level Visual Content Descriptor combining visual and textual cues as collaterals

In this paper, we introduce a novel high-level visual content descriptor which is devised for performing semantic-based image classification and retrieval. The work can be treated as an attempt to bridge the so called “semantic gap”. The proposed image feature vector model is fundamentally underpinned by the image labelling framework, called Collaterally Confirmed Labelling (CCL), which incorporates the collateral knowledge extracted from the collateral texts of the images with the state-of-the-art low-level image processing and visual feature extraction techniques for automatically assigning linguistic keywords to image regions. Two different high-level image feature vector models are developed based on the CCL labelling of results for the purposes of image data clustering and retrieval respectively. A subset of the Corel image collection has been used for evaluating our proposed method. The experimental results to-date already indicates that our proposed semantic-based visual content descriptors outperform both traditional visual and textual image feature models.

Fusion of motion segmentation algorithms

Many algorithms have been developed to achieve motion segmentation for video surveillance. The algorithms produce varying performances under the infinite amount of changing conditions. It has been recognised that individually these algorithms have useful properties. Fusing the statistical result of these algorithms is investigated, with robust motion segmentation in mind.