This paper investigates the problems of outliers and/or noise in surface segmentation and proposes a statistically robust segmentation algorithm for laser scanning 3-D point cloud data. Principal component analysis (PCA)-based local saliency features, e.g., normal and curvature, have been frequently used in many ways for point cloud segmentation. However, PCA is sensitive to outliers; saliency features from PCA are nonrobust and inaccurate in the presence of outliers; consequently, segmentation results can be erroneous and unreliable. As a remedy, robust techniques, e.g., RANdom SAmple Consensus (RANSAC), and/or robust versions of PCA (RPCA) have been proposed. However, RANSAC is influenced by the well-known swamping effect, and RPCA methods are computationally intensive for point cloud processing. We propose a region growing based robust segmentation algorithm that uses a recently introduced maximum consistency with minimum distance based robust diagnostic PCA (RDPCA) approach to get robust saliency features. Experiments using synthetic and laser scanning data sets show that the RDPCA-based method has an intrinsic ability to deal with outlier- and/or noise-contaminated data. Results for a synthetic data set show that RDPCA is 105 times faster than RPCA and gives more accurate and robust results when compared with other segmentation methods. Compared with RANSAC and RPCA based methods, RDPCA takes almost the same time as RANSAC, but RANSAC results are markedly worse than RPCA and RDPCA results. Coupled with a segment merging algorithm, the proposed method is efficient for huge volumes of point cloud data consisting of complex objects surfaces from mobile, terrestrial, and aerial laser scanning systems.

This paper introduces robust algorithms for extracting the ground points in laser scanning 3-D point cloud data. Global polynomial functions have been used for filtering algorithms for point cloud data; however, it is not suitable as it may lead to bias for the filtering algorithms and can cause misclassification errors when many different objects are present. In this paper, robust statistical approaches are coupled with locally weighted 2-D regression that fits without any predefined global function for the variables of interest. Algorithms are performed iteratively on 2-D profiles: x - z and y - z. The z (elevation) values are robustly down weighted based on the residuals for the fitted points. The new set of down-weighted z values, along with the corresponding x (or y) values, is used to get a new fit for the lower surface level. The process of fitting and down weighting continues until the difference between two consecutive fits is insignificant. The final fit is the required ground level, and the ground surface points are those that fall within the ground level and the level after adding some threshold value with the ground level for z values. Experimental results are compared with the recently proposed segmentation method through simulated and real mobile laser scanning point clouds from urban areas that include many objects that appear in road scenes such as short walls, large buildings, electric poles, signposts, and cars. Results show that the proposed robust methods efficiently extract ground surface points with better than 97% accuracy.

Vehicle-based mobile mapping systems capture co-registered imagery and 3D point cloud information over hundreds of kilometres of transport corridor. Methods for extracting information from these large datasets are labour intensive and automatic methods are desired. In addition, such methods need to be easily configured by non-expert users to detect and measure many classes of objects. This paper describes a workflow to take a large number of image and depth features, use machine learning to generate an object detection system that is fast to configure and run. The output is high detection of the objects of interest but with an acceptable number of false alarms. This is desirable as the output is fed into a more complex and hence more computationally expensive analysis system to reject the false alarms and measure the remaining objects. Image and depth features from bounding boxes around objects of interest and random background are used for training with some popular learning algorithms. The interface allows a non-expert user to observe the performance and make modifications to improve the performance. Copyright © 2014 SCITEPRESS.

Spatial Data Infrastructures have recently become a crucial part of national infrastructures. Example users are governments using them to make informed policy decisions and the private sector using them in order to understand their customers better. It is estimated that Australian spatial industry revenue is in excess of $1.35 billion annually. The Cooperative Research Centre for Spatial Information in Australia is currently working with the Commonwealth, jurisdictions and the private sector to understand the research required in this area to support both the public and private sectors in their decision and policy making based upon these infrastructures. This paper presents an early perspective as to possible research areas in this field. An underlying theme that reoccurs in the research is the need to consider usability of such systems and the need to move beyond just data to orchestration of processes to obtain derived products.

The timetabling problem is to find a timetable solution by assigning time and resources to sessions that satisfy a set of constraints. Traditionally, research has focused on optimization towards a final solution but this paper focuses on minimizing disturbance impact due to changing conditions. A Multi-Agent System (MAS) is proposed in which users are represented as autonomous agents negotiating with one another to repair a timetable. From repeated negotiations, agents learn to develop a model of other agent's preferences. The MAS is simulated on a factorial experiment set up and varying the cooperation level, learning model and selection strategy. A provenance-centred approach is adopted to improve the human aspect of timetabling to allow users to derive the steps towards a solution and make changes to influence the solution.

Despite the growth of stereoscopic technology, there has been very little research done into the quantitative benefits of 3D over 2D visualisation systems. This study seeks to quantify and compare the human brain's response to 2D versus 3D images using EEG technology, in particular focusing on the trend in changes between the p300 response resulting from 2D and 3D stimulus sequences using the oddball paradigm. It was found that the p300 response to the 3D stimuli degraded at a faster rate than the 2D stimuli as target spotting task difficulty was raised.

This paper presents a method for robustly stabilising omnidirectional video given the presence of significant rotations and translations by creating a virtual camera and using a combination of sensor fusion and scene tracking. Real time rotational movements of the camera are measured by an Inertial Measurement Unit (IMU), which provides an initial estimate of the ego-motion of the camera platform. Image registration is then used to refine these estimates. The calculated ego-motion is then used to adjust an extract of the omnidirectional video, forming a virtual camera that is focused on the scene. Experiments show the technique is effective under challenging ego-motions and overcomes deficiencies that are associated with unimodal approaches making it robust and suitable to be used in many surveillance applications.

The majority of multi-class pattern classification techniques are proposed for learning from balanced datasets. However, in several real-world domains, the datasets have imbalanced data distribution, where some classes of data may have few training examples compared for other classes. In this paper we present our research in learning from imbalanced multi-class data and propose a new approach, named Multi-IM, to deal with this problem. Multi-IM derives its fundamentals from the probabilistic relational technique (PRMs-IM), designed for learning from imbalanced relational data for the two-class problem. Multi-IM extends PRMs-IM to a generalized framework for multi-class imbalanced learning for both relational and non-relational domains.

This paper proposes a number of methods to evaluate features in the context of people tracking in complex environments. Such environments will have varying lighting conditions (the subject of this paper), occlusions by other people and objects, as well as a varying number of people. The paper concentrates on edge features because of their insensitivity to changes in illumination and camera movements. It assumes that some form of model-based processing will be used for recognition and tracking so as to be able to deal with partially visible people. This requires the adaptive choice of what parts of people need to be tracked using the best combination of features. A number of measures are proposed to quantify edge performance that are illustrated for a number of edge detectors on a number of video sequences that have different properties or contexts.

Privacy is an important aspect of pervasive and ubiquitous computing systems, and, in particular, pervasive healthcare. With reference to previous approaches on developing privacy sensitive pervasive healthcare applications, we detail a framework for the design of such systems that aims to minimise the impact of privacy on such systems. In reviewing previous approaches, we extract and combine common elements in order to unify the approaches and create a more formal methodology for designing privacy mechanisms in pervasive healthcare applications. In doing so we also consider the manner in which ubiquitous technologies impact on privacy and methods for reducing this impact. We demonstrate how the framework can be applied by using examples from the previous approaches. In addressing privacy issues, the framework aims to remove a large obstacle to deployment of pervasive healthcare systems, acceptance of the technology.

Traditional learning techniques learn from flat data files with the assumption that each class has a similar number of examples. However, the majority of real-world data are stored as relational systems with imbalanced data distribution, where one class of data is over-represented as compared with other classes. We propose to extend a relational learning technique called Probabilistic Relational Models (PRMs) to deal with the imbalanced class problem. We address learning from imbalanced relational data using an ensemble of PRMs and propose a new model: the PRMs-IM. We show the performance of PRMs-IM on a real university relational database to identify students at risk.

In this paper we present preliminary work implementing dynamic privacy in public surveillance. The aim is to maximise the privacy of those under surveillance, while giving an observer access to sufficient information to perform their duties. As these aspects are in conflict, a dynamic approach to privacy is required to balance the systempsilas purpose with the systempsilas privacy. Dynamic privacy is achieved by accounting for the situation, or context, within the environment. The context is determined by a number of visual features that are combined and then used to determine an appropriate level of privacy.

Soil moisture is an important variable that controls the partition of rainfall into infiltration and run-off. This plays an important role in the prediction of erosion, flood or drought. Passive microwave remote sensing data has great potential for providing estimates of soil moisture. This is mainly due to the minimal weather influence on passive microwave data and its ability to penetrate through clouds. The artificial neural network (ANN) is a method that tries simulates human intelligence by crudely imitating the way a human brain learns. This method has been especially useful for mapping non-linear and ill-posed problems. Soil moisture retrieval is an example of a non-linear problem. An explicit inverse of the physical process can be built using an ANN to map the passive microwave measurements into land surface parameters such as soil moisture. For this paper, the ANN method used to create the explicit inverse function is divided into (i.) single parameter retrieval of the soil moisture value given the passive microwave measurements, and (ii.) multi-parameter retrieval of a number of land surface parameters, i.e. soil temperature, surface roughness, together with soil moisture value given the passive microwave measurements. This paper examines these methods in the context of retrieving surface soil moisture values given microwave radiometric data and discusses key issues that will need to be addressed to improve mapping performance and to produce operational systems.

The accuracy of an Artificial Neural Network (ANN) depends on the representativeness of the data used to train it. Although it is known that an ANN will function well as long as the pattern of the input data is similar to the testing data, there has been no research on the effect of data "similarity" on the accuracy of the network outputs. In this paper, an ANN model is used to retrieve soil moisture from the H- and V-polarized brightness temperature obtained. The research discussed in this paper is focused on the standard deviation of the data used for training and testing of the ANN. It is shown that similarity in standard deviation is a good indicator to choose representative training and testing data set. By doing this, the accuracy of retrieval increases from around 22% volume/volume (v/v) of Root Mean Square Error (RMSE) to around 2%(v/v).

Surveillance applications in private environments such as smart houses require a privacy management policy if such systems are to be accepted by the occupants of the environment. This is due to the invasive nature of surveillance, and the private nature of the home. In this article, we propose a framework for dynamically altering the privacy policy applied to the monitoring of a smart house based on the situation within the environment. Initially the situation, or context, within the environment is determined; we identify several factors for determining environmental context, and propose methods to quantify the context using audio and binary sensor data. The context is then mapped to an appropriate privacy policy, which is implemented by applying data hiding techniques to control access to data gathered from various information sources. The significance of this work lies in the examination of privacy issues related to assisted-living smart house environments. A single privacy policy in such applications would be either too restrictive for an observer, for example, a carer, or too invasive for the occupants. We address this by proposing a dynamic method, with the aim of decreasing the invasiveness of the technology, while retaining the purpose of the system.

This paper addresses the problem of markerless tracking of a human in full 3D with a high-dimensional (29D) body model. Most work in this area has been focused on achieving accurate tracking in order to replace marker-based motion capture, but do so at the cost of relying on relatively clean observing conditions. This paper takes a different perspective, proposing a body-tracking model that is explicitly designed to handle real-world conditions such as occlusions by scene objects, failure recovery, long-term tracking, auto-initialisation, generalisation to different people and integration with action recognition. To achieve these goals, an action's motions are modelled with a variant of the hierarchical hidden Markov model. The model is quantitatively evaluated with several tests, including comparison to the annealed particle filter, tracking different people and tracking with a reduced resolution and frame rate.

A smart house can be regarded as a surveillance environment in which the person being observed carries out activities that range from intimate to more public. What can be observed depends on the activity, the person observing (e.g. a carer) and policy. In assisted living smart house environments, a single privacy policy, applied throughout, would be either too invasive for an occupant, or too restrictive for an observer, due to the conflicting goals of surveillance and private environments. Hence, we propose a dynamic method for altering the level of privacy in the environment based on the context, the situation within the environment, encompassing factors relevant to ensuring the occupant's safety and privacy. The context is mapped to an appropriate level of privacy, which is implemented by controlling access to data sources (e.g. video) using data hiding techniques. The aim of this work is to decrease the invasiveness of the technology, while retaining the purpose of the system.

In this paper we present a framework for addressing privacy issues raised by the monitoring of assisted living smart house environments. In home environments, the conflict between the goals of the surveillance, and the private nature of the home, raises the issue of occupant privacy. This issue needs to be addressed if applications are to be accepted by the occupant. We identify four key properties required for the design of privacy sensitive ubiquitous computing applications. Subsequently, we develop a dynamic and flexible method for implementing privacy measures through controlling access to data, and an interface to provide feedback to the occupant, enabling them to control the implemented privacy measures. We form a generic framework for implementing privacy sensitive ubiquitous computing applications based on previous applications within the field. This framework was then extended and used to develop a specific framework for a privacy sensitive smart house. The approach proposed in the framework dynamically applies privacy measures to multi-modal data according to the situation, or context, of the environment. We further test an implementation of the privacy measures, and detail methods to implement feedback and control. The approach aims to decrease the invasiveness of the surveillance, while retaining the purpose of the assisted living environment.

We present a method for foreground/background separation of audio using a background modelling technique. The technique models the background in an online, unsupervised, and adaptive fashion, and is designed for application to long term surveillance and monitoring problems. The background is determined using a statistical method to model the states of the audio over time. In addition, three methods are used to increase the accuracy of background modelling in complex audio environments. Such environments can cause the failure of the statistical model to accurately capture the background states. An entropy-based approach is used to unify background representations fragmented over multiple states of the statistical model. The approach successfully unifies such background states, resulting in a more robust background model. We adaptively adjust the number of states considered background according to background complexity, resulting in the more accurate classification of background models. Finally, we use an auxiliary model cache to retain potential background states in the system. This prevents the deletion of such states due to a rapid influx of observed states that can occur for highly dynamic sections of the audio signal. The separation algorithm was successfully applied to a number of audio environments representing monitoring applications.

This paper describes techniques used in a Web-based decision support system to enhance dermatological diagnosis for rural and remote communities. We proposed a technique to improve the query representation and shorten the length of the query process, and a knowledge validation technique to help the domain expert supervise and maintain a consistent knowledge base. Experiments were conducted to test the effectiveness of each technique. Results indicated that the new querying technique improves a decision tree query approach. In addition, the results showed that supervised knowledge updating can improve the identification and handling of ambiguous cases to produce 100% correct classification.

This paper describes an approach to representing normal activities in a smart house based on the concept of anxiety. Anxiety is computed as a function of time and is kept low by interactions of an occupant with the various devices in a house. Abnormality is indicated by a lack of activity or the wrong activity which will cause anxiety to rise ultimately raising an alarm, querying the occupant and/or alerting a carer in real-time. Anxiety is formulated using probabilistic models that describe how people interact with devices in combinations. These models can be learnt interactively as the smart house acts pessimistically enquiring of the occupant if what they are doing is normal. Results are presented for a number of kitchen scenarios and for different formulations of anxiety.

Traditional methods of object recognition are reliant on shape and so are very difficult to apply in cluttered, wide angle and low detail views such as surveillance scenes. To address this, a method of indirect object recognition is proposed, where human activity is used to infer both the location and identity of objects. No shape analysis is necessary. The concept is dubbed 'interaction signatures', since the premise is that a human interacts with objects in ways characteristic of the function of that object - for example, a person sits in a chair and drinks from a cup. The human-centred approach means that recognition is possible in low detail views and is largely invariant to the shape of objects within the same functional class. This paper implements a Bayesian network for classifying region patches with object labels, building upon our previous work in automatically segmenting and recognising a human's interactions with the objects. Experiments show that interaction signatures can successfully find and label objects in low detail views and are equally effective at recognising test objects that differ markedly in appearance from the training objects.

This paper is concerned with modelling background audio online to detect foreground sounds in complex audio environments for surveillance and smart home applications. We examine and expand upon previous work in the audio and video domains, and propose a new implementation of an audio background modelling algorithm, addressing the complexities of audio data. A number of audio features characterizing different aspects of the audio content were analysed to determine the factors relevant to the determination of the background audio. We test the algorithms on three audio data sets of varying complexity. The new approach was successful in modelling the background audio for the test data.

We present results on the recognition of intentional human gestures for video annotation and retrieval. We define a gesture as a particular, repeatable, human movement having a predefined meaning. An obvious application of the work is in sports video annotation where umpire gestures indicate specific events. Our approach is to augment video with data obtained from accelerometers worn as wrist bands by one or more officials. We present the recognition performance using a Hidden Markov Model approach for gesture modeling with both isolated gestures and gestures segmented from a stream.

The recognition of activities from sensory data is important in advanced surveillance systems to enable prediction of high-level goals and intentions of the target under surveillance. The problem is complicated by sensory noise and complex activity spanning large spatial and temporal extents. The paper presents a system for recognizing high-level human activities from multi-camera video data in complex spatial environments. The Abstract Hidden Markov mEmory Model (AHMEM) is used to deal with noise and scalability. The AHMEM is an extension of the Abstract Hidden Markov Model (AHMM) that allows us to represent a richer class of both state-dependent and context-free behaviors. The model also supports integration with low-level sensory models and efficient probabilistic inference. We present experimental results showing the ability of the system to perform real-time monitoring and recognition of complex behaviors of people from observing their trajectories within a real, complex indoor environment.

Seeking to extend the functional capability of the elderly, we explore the use of probabilistic methods to learn and recognise human activity in order to provide monitoring support. We propose a novel approach to learning the hierarchical structure of sequences of human actions through the application of the hierarchical hidden Markov model (HHMM). Experimental results are presented for learning and recognising sequences of typical activities in a home.

The paper presents a method for finding and classifying objects within real-world scenes by using the activity of humans interacting with these objects to infer the object's identity. Objects are labelled using evidence accumulated over time and multiple instances of human interactions. This approach is inspired by the problems and opportunities that exist in recognition tasks for intelligent homes, namely cluttered, wide-angle views coupled with significant and repeated human activity within the scene. The advantages of such an approach include the ability to detect salient objects in a cluttered scene, independent of the object's physical structure, adapt to changes in the scene and resolve conflicts in labels by weight of past evidence. This initial investigation seeks to label chairs and open floor spaces by recognising activities such as walking and silting. Findings show that the approach can locate objects with a reasonably high degree of accuracy, with occlusions of the human actor being a significant aid in reducing over-labelling.

This paper describes an application of camera motion estimation to index cricket games. The shots are labeled with the type of shot: glance left, glance right, left drive, right drive, left cut, right pull and straight drive. The method has the advantages that it is fast and avoids complex image segmentation. The classification of the cricket shots is done using an incremental learning algorithm. We tested the method on over 600 shots and the results show that the system has a classification accuracy of 74%.

In this paper, we present a method for recognising an agent's behaviour in dynamic, noisy, uncertain domains, and across multiple levels of abstraction. We term this problem on-line plan recognition under uncertainty and view it generally as probabilistic inference on the stochastic process representing the execution of the agent's plan. Our contributions in this paper are twofold. In terms of probabilistic inference, we introduce the Abstract Hidden Markov Model (AHMM), a novel type of stochastic processes, provide its dynamic Bayesian network (DBN) structure and analyse the properties of this network. We then describe an application of the Rao-Blackwellised Particle Filter to the AHMM which allows us to construct an efficient, hybrid inference method for this model. In terms of plan recognition, we propose a novel plan recognition framework based on the AHMM as the plan execution model. The Rao-Blackwellised hybrid inference for AHMM can take advantage of the independence properties inherent in a model of plan execution, leading to an algorithm for online probabilistic plan recognition that scales well with the number of levels in the plan hierarchy. This illustrates that while stochastic models for plan execution can be complex, they exhibit special structures which, if exploited, can lead to efficient plan recognition algorithms. We demonstrate the usefulness of the AHMM framework via a behaviour recognition system in a complex spatial environment using distributed video surveillance data.

In this paper, we consider the problem of tracking an object and predicting the object's future trajectory in a wide-area environment, with complex spatial layout and the use of multiple sensors/cameras. To solve this problem, there is a need for representing the dynamic and noisy data in the tracking tasks, and dealing with them at different levels of detail. We employ the Abstract Hidden Markov Models (AHMM), an extension of the well-known Hidden Markov Model (HMM) and a special type of Dynamic Probabilistic Network (DPN), as our underlying representation framework. The AHMM allows us to explicitly encode the hierarchy of connected spatial locations, making it scalable to the size of the environment being modeled. We describe an application for tracking human movement in an office-like spatial layout where the AHMM is used to track and predict the evolution of object trajectories at different levels of detail.

Surveillance in wide-area spatial environments is characterised by complex spatial layouts, large state space, and the use of multiple cameras/sensors. To solve this problem, there is a need for representing the dynamic and noisy data in the tracking tasks, and dealing with them at different levels of detail. This requirement is particularly suited to the layered dynamic probabilistic network (LDPN), a special type of dynamic probabilistic network. In this paper, we propose the use of LDPN as the integrated framework for tracking in wide-area environments. We illustrate, with the help of a synthetic tracking scenario, how the parameters of the LDPN can be estimated from training data, and then used to draw predictions and answer queries about unseen tracks at various levels of detail.

This paper presents a new neural network architecture based on the prior work on adaptive resonance theory (ART) that is capable of incremental learning. The architecture, called IFOSART, addresses several problems encountered in previous incremental learning ART networks while maintaining their advantages. The paper addresses the plasticity-stability issue by incorporating a concept of time into the network. Also addressed is the problem of handling exceptions in the data, as opposed to data simplification and generalization. IFOSART incorporates a noise removal mechanism to prevent the excessive proliferation of exception categories and to remove any instantiated noise categories from the network. Results of experiments undertaken show that the presented network is comparable in its generalization ability to other types of neural, fuzzy and traditional classifiers while maintaining a fair tolerance towards noise.

We discuss combining incremental learning and incremental recognition to classify patterns consisting of multiple objects, each represented by multiple spatio-temporal features. Importantly the technique allows for ambiguity in terms of the positions of the start and finish of the pattern. This involves a progressive classification which considers the data at each time instance in the query and thus provides a probable answer before all the query information becomes available. We present two methods that combine incremental learning and incremental recognition: a time instance method and an overall best match method.

Describes results of using different features for the pose refinement of 3D objects i.e. find the best transformation between a 3D model and a 2D view of the object. This is a six degrees of freedom problem which is seen as essential to solve for 3D object recognition. Previous research has investigated various matching measure for pose refinement using edge points from the model and all edge pixels in the image, in the presence of noise and other objects. This paper extends the work to consider other features such as vertices and lines giving results to show the usefulness of each feature.

This paper describes a low-cost interactive active monocular range finder and illustrates the effect of introducing interactivity to the range acquisition process. The range finder consists of only one camera and a laser pointer to which three LEDs are attached. When a user scans the laser along surfaces of objects, the camera captures the image of spots (one from the laser and the others from LEDs), and triangulation is carried our using the camera's viewing direction and the optical axis of the laser. The user interaction allows the range finder to acquire range data in which the sampling rate varies across the object depending on the underlying surface structures. Moreover the processes of separating objects from the background and/or finding parts in the object can be achieved using the operator's knowledge of the objects.

This paper describes object-centered symbolic representation and distributed matching strategies of 3D objects in a schematic form which occur in engineering drawings and maps. The object-centered representation has a hierarchical structure and is constructed from symbolic representations of schematics. With this representation, two independent schematics representing the same object can be matched. We also consider matching strategies using distributed algorithms. The object recognition is carried out with two matching methods: (1) matching between an object model and observed data at the lowest level of the hierarchy, and (2) constraints propagation. The first is carried out with symbolic Hopfield-type neural networks and the second is achieved via hierarchical winner-takes-all algorithms.

In applications such as tracking and surveillance in large spatial environments, there is a need for representing dynamic and noisy data and at the same time dealing with them at different levels of detail. In the spatial domain, there has been work dealing with these two issues separately, however, there is no existing common framework for dealing with both of them. In this paper, we propose a new representation framework called the Layered Dynamic Probabilistic Network (LDPN), a special type of Dynamic Probabilistic Network (DPN), capable of handling uncertainty and representing spatial data at various levels of detail. The framework is thus particularly suited to applications in wide-area environments which are characterised by large region size, complex spatial layout and multiple sensors/cameras. For example, a building has three levels: entry/exit to the building, entry/exit between rooms and moving within rooms. To avoid the problem of a relatively large state space associated with a large spatial environment, the LDPN explicitly encodes the hierarchy of connected spatial locations, making it scalable to the size of the environment being modelled. There are three main advantages of the LDPN. First, the reduction in state space makes it suitable for dealing with wide area surveillance involving multiple sensors. Second, it offers a hierarchy of intervals for indexing temporal data. Lastly, the explicit representation of intermediate sub-goals allows for the extension of the framework to easily represent group interactions by allowing coupling between sub-goal layers of different individuals or objects. We describe an adaptation of the likelihood sampling inference scheme for the LDPN, and illustrate its use in a hypothetical surveillance scenario.

We describe a color image transform which is capable of encoding spatio-chromatic image features. The method relies on the development of physical interpretations of complex image Fourier transforms allowing the direct extraction of spatio-chromatic structures from images. The process is analysed using an elliptical representation in order to determine filter characteristics and responses. Both a spatial and chromatic filter are developed and used on test images with results showing the ease with which this technique can be utilised.

We outline an approach to classifying and detecting behaviours from surveillance data. Simple pairwise movement patterns are learned and used as building blocks to describe behaviour over a temporal sequence, or compared with other pairs to detect group behaviour. As the pair primitives are easy to redefine and learn, and complex behaviour over time is specified by the user as a sequence of pair primitives, this approach provides a flexible yet robust method of detecting complex movement in a wide variety of domains.

We explore the use of natural language understanding and image processing to index and query American Football tapes. We present a model for representing spatio-temporal characteristics of multiple objects in dynamic scenes in this domain, and a recognition system which uses the model to recognise American Football plays.

This paper is concerned with the design and manufacturing of 2D and 3D components in the context of process modeling and inspection. A methodology is proposed that models manufacturing processes and learns shape deformations that are the result of out-of-tolerance components. Then during inspection, a deformed shape is analyzed and the process parameter values determined. The key aspect of this approach is that the mapping between process parameters and shape is determined which is not possible using standard process modeling approaches such as finite element analysis. This is different to the traditional approach to tolerancing which is very much as additional procedure to design and doesn't consider process parameter variation, just shape variation. A review of tolerancing is presented as well as a description of techniques for process modeling.

A popular paradigm in computer vision is based on dividing the vision problem into three stages namely segmentation, feature extraction and recognition. For example edge detection followed by line detection followed by planar object recognition. It can be argued that each of these stages needs to be thoroughly described to enable vision systems to be configured with predictable performance. However an alternative view is that the performance of each stage is not in itself important as long as the overall performance is acceptable. This paper discusses feature performance concentrating on the assessmentof edge-based feature detection and object recognition. Evaluation techniques are discussed for assessing arc and line detection algorithmsand for features in the context of verification and pose refinement strategies. These techniques can then be used for the design and integration of indexing and verification stages of object recognition. A theme of the paper is the need to assess feature extraction in the context of the chosen task.

Abstract — We discuss the advantages and disadvantages of two approaches to model selection: the information theo-retic method suggested by Kanatani [3] and others, and our heuristic sequential selection method [9].

Log polar transformations for space variant systems have been proposed and used in active vision research. The idea is to generate an image with a varying resolution over a wide angle field of view. The fovea is of high resolution and the periphery is of exponentially reduced resolution. The justifications for such a sensor are: (i) it provides high resolution and a wide viewing angle, (ii) feature invariance in the fovea simplifies foveation, and (iii) it allows multiresolution analysis. The receptor density of the human retina is very high, i.e. of the order of 10/sup 6/ receptors at the fovea. The question is, what resolution should space variant active vision systems have? Real visual sensors have been implemented but is the resolution produced high enough? This paper investigates the resolution requirements of a space variant sensor by simulation for a tracking system using raytracing.

Without the ability to foveate on and maintain foveation, active vision for applications such as surveillance, object recognition and object tracking are difficult to build. Although foveation in cartesian coordinates is being actively pursued by many, multi-resolution high accuracy foveation in log polar space has not been given much attention. This paper addresses the use of foveation to track a single object as well as multiple objects for a simulated space variant active vision system. Complex logarithmic mapping is chosen firstly because it provides high resolution and wide angle viewing. Secondly, the spatially variant structure of log polar space leads to an object increasing in size as it moves towards the fovea. This is important as we know which object is closer to the fovea at any instant in time.

Object recognition can be defined as consisting of two main stages: indexing and verification, Indexing has received much attention in the literature with many schemes developed. However verification has received less attention and is generally used in its simplest form. This paper discusses evaluation techniques for assessing features in the context of verification and pose refinement strategies. Two metrics are considered: a typical Euclidean metric and the Hausdorff metric which is attracting interest in the vision community. These techniques can be used for the design and integration of indexing and verification stages of object recognition.

This paper describes and demonstrates the operation and performance of an algorithm for segmenting connected points into a combination of representations such as lines, circular, elliptical and superelliptical arcs, and polynomials. The algorithm has a number of interesting properties including being scale invariant, nonparametric, general purpose, and efficient.

This paper describes an investigation into the use of parametric 2D models describing the movement of edges for the determination of possible 3D shape and hence function of an object. An assumption of this research is that the camera can foveate and track particular features. It is argued that simple 2D analytic descriptions of the movement of edges can infer 3D shape while the camera is moved. This uses an advantage of foveation i.e. The problem becomes object centred. The problem of correspondence for numerous edge points is overcome by the use of a tree based representation for the competing hypotheses. Numerous hypothesis are maintained simultaneously and it does not rely on a single kinematic model which assumes constant velocity or acceleration. The numerous advantages of this strategy are described.>

In this paper we describe a new technique for ID and 2D edge feature extraction to subpixel accuracy using edge models and the local energy approach. A candidate edge is modeled as one of a number of parametric edge models, and the fit is refined by a least-squared error fitting technique.>

The distance transform has been proposed for use in computer vision for a number of applications such as matching and skeletonisation. This paper proposes two things: (1) a multi-scale distance transform to overcome the need to choose edge thresholds and scale and (2) the addition of various saliency factors such as edge strength, length and curvature to the basic distance transform to improve its effectiveness. Results are presented for applications of matching and snake fitting. 1

Ellipses seen in an image may be the 2-D projection of 3-D circles from the scene. Given this assumption, ellipses can be grouped into perceptual groups from which inferences about the 3-D structure of objects can be made. Methods are proposed for extracting groupings corresponding to surfaces of revolution. A Hough transform approach is used for grouping, after which the confidence in the plausibility of the perceptual group is improved by detecting symmetry groupings.>

Theories and techniques of shape constancy and shape from contour often assert that ellipses seen in an image are the 2D projection of 3D circles from the scene. Given this assumption, several methods for grouping ellipses are described. From these perceptual groups inferences about the 3D structure of their generating circles are made. This will be useful for model invokation. The world (both natural and man-made) is constrained by physical laws to a finite number of basic patterns. Certain three dimensional relationships between features in a scene give rise to two dimensional relationships that are relatively viewpoint invariant and can be distinguished as non-accidental. Thus, those regularities observable in the image arise not by accident, but are projections of real regularities in the scene. The Gestalt movement proposed

A method is described for segmenting edge data into a combination of straight lines and elliptic arcs. The two-stage process first segments the data into straight line segments. Ellipses are then fitted to the line data. This is much faster than curve fitting directly to pixel data since the lines provide a great reduction in data. Segmentation is performed in the paradigm suggested by D.G. Lowe (1987). A measure of significance is defined that produces a scale-invariant description and allows the replacement of sequences of line segments by ellipses without requiring any thresholds. A method for fitting ellipses to arbitrary curves, essential for this algorithm, has been developed, based on an iterative Kalman filter. This is guaranteed to produce an elliptical fit even though the best conic fit may be a hyperbola or parabola.>

A system for automatic inspection of bare printed circuit boards is described that uses three concepts: multiple inspection systems; learn and inspect phases of inspection; a hierarchy of detection processes to satisfy the four basic criteria for a successful inspection system, namely detection performance, speed, cost, and flexibility. The system has reached a stage at which industrial implementation can be considered.