Feature selection using expected attainable discrimination

We propose expected attainable discrimination (EAD) as a measure to select discrete valued features for reliable discrimination between two classes of data. EAD is an average of the area under the ROC curves obtained when a simple histogram probability density model is trained and tested on many random partitions of a data set. EAD can be incorporated into various stepwise search methods to determine promising subsets of features, particularly when misclassification costs are difficult or impossible to specify. Experimental application to the problem of risk prediction in pregnancy is described.

Long-range stereo visual odometry for unmanned aerial vehicles

This thesis explored the utility of long-range stereo visual odometry for application on Unmanned Aerial Vehicles. Novel parameterisations and initialisation routines were developed for the long-range case of stereo visual odometry and new optimisation techniques were implemented to improve the robustness of visual odometry in this difficult scenario. In doing so, the applications of stereo visual odometry were expanded and shown to perform adequately in situations that were previously unworkable.

Functional MRI of facial emotion recognition deficits in schizophrenia and their electrophysiological correlates

Empirical evidence suggests impaired facial emotion recognition in schizophrenia. However, the nature of this deficit is the subject of ongoing research. The current study tested the hypothesis that a generalized deficit at an early stage of face-specific processing (i.e. putatively subserved by the fusiform gyrus) accounts for impaired facial emotion recognition in schizophrenia as opposed to the Negative Emotion-specific Deficit Model, which suggests impaired facial information processing at subsequent stages. Event-related potentials (ERPs) were recorded from 11 schizophrenia patients and 15 matched controls while performing a gender discrimination and a facial emotion recognition task. Significant reduction of the face-specific vertex positive potential (VPP) at a peak latency of 165 ms was confirmed in schizophrenia subjects whereas their early visual processing, as indexed by P1, was found to be intact. Attenuated VPP was found to correlate with subsequent P3 amplitude reduction and to predict accuracy when performing a facial emotion discrimination task. A subset of ten schizophrenia patients and ten matched healthy control subjects also performed similar tasks in the magnetic resonance imaging scanner. Patients showed reduced blood oxygenation level-dependent (BOLD) activation in the fusiform, inferior frontal, middle temporal and middle occipital gyrus as well as in the amygdala. Correlation analyses revealed that VPP and the subsequent P3a ERP components predict fusiform gyrus BOLD activation. These results suggest that problems in facial affect recognition in schizophrenia may represent flow-on effects of a generalized deficit in early visual processing.

Visual control for automated aircraft collision avoidance systems

This thesis presents a new vision-based decision and control strategy for automated aircraft collision avoidance that can be realistically applied to the See and Avoid problem. The effectiveness of the control strategy positions the research as a major contribution toward realising the simultaneous operation of manned and unmanned aircraft within civilian airspace. Key developments include novel classical and visual predictive control frameworks, and a performance evaluation technique aligned with existing aviation practise and applicable to autonomous systems. The overall approach is demonstrated through experimental results on a small multirotor unmanned aircraft, and through high fidelity probabilistic simulation studies.

A neuronal topography of visual and acoustic fear conditioning within the amygdala

The lateral amygdala (LA) receives information from auditory and visual sensory modalities, and uses this information to encode lasting memories that predict threat. One unresolved question about the amygdala is how multiple memories, derived from different sensory modalities, are organized at the level of neuronal ensembles. We previously showed that fear conditioning using an auditory conditioned stimulus (CS) was spatially allocated to a stable topography of neurons within the dorsolateral amygdala (LAd) (Bergstrom et al, 2011). Here, we asked how fear conditioning using a visual CS is topographically organized within the amygdala. To induce a lasting fear memory trace we paired either an auditory (2 khz, 55 dB, 20 s) or visual (1 Hz, 0.5 s on/0.5 s off, 35 lux, 20 s) CS with a mild foot shock unconditioned stimulus (0.6 mA, 0.5 s). To detect learning-induced plasticity in amygdala neurons, we used immunohistochemistry with an antibody for phosphorylated mitogen-activated protein kinase (pMAPK). Using a principal components analysis-based approach to extract and visualize spatial patterns, we uncovered two unique spatial patterns of activated neurons in the LA that were associated with auditory and visual fear conditioning. The first spatial pattern was specific to auditory cued fear conditioning and consisted of activated neurons topographically organized throughout the LAd and ventrolateral nuclei (LAvl) of the LA. The second spatial pattern overlapped for auditory and visual fear conditioning and was comprised of activated neurons located mainly within the LAvl. Overall, the density of pMAPK labeled cells throughout the LA was greatest in the auditory CS group, even though freezing in response to the visual and auditory CS was equivalent. There were no differences detected in the number of pMAPK activated neurons within the basal amygdala nuclei. Together, these results provide the first basic knowledge about the organizational structure of two different fear engrams within the amygdala and suggest they are dissociable at the level of neuronal ensembles within the LA

IMU aided 3D visual odometry for car-like vehicles

We present a method for calculating odome- try in three-dimensions for car-like ground ve- hicles with an Ackerman-like steering model. In our approach we use the information from a single camera to derive the odometry in the plane and fuse it with roll and pitch informa- tion derived from an on-board IMU to extend to three-dimensions, thus providing odometric altitude as well as traditional x and y transla- tion. We have mounted the odometry module on a standard Toyota Prado SUV and present results from a car-park environment as well as from an off-road track.

Supervised latent dirichlet allocation models for efficient activity representation

Local spatio-temporal features with a Bag-of-visual words model is a popular approach used in human action recognition. Bag-of-features methods suffer from several challenges such as extracting appropriate appearance and motion features from videos, converting extracted features appropriate for classification and designing a suitable classification framework. In this paper we address the problem of efficiently representing the extracted features for classification to improve the overall performance. We introduce two generative supervised topic models, maximum entropy discrimination LDA (MedLDA) and class- specific simplex LDA (css-LDA), to encode the raw features suitable for discriminative SVM based classification. Unsupervised LDA models disconnect topic discovery from the classification task, hence yield poor results compared to the baseline Bag-of-words framework. On the other hand supervised LDA techniques learn the topic structure by considering the class labels and improve the recognition accuracy significantly. MedLDA maximizes likelihood and within class margins using max-margin techniques and yields a sparse highly discriminative topic structure; while in css-LDA separate class specific topics are learned instead of common set of topics across the entire dataset. In our representation first topics are learned and then each video is represented as a topic proportion vector, i.e. it can be comparable to a histogram of topics. Finally SVM classification is done on the learned topic proportion vector. We demonstrate the efficiency of the above two representation techniques through the experiments carried out in two popular datasets. Experimental results demonstrate significantly improved performance compared to the baseline Bag-of-features framework which uses kmeans to construct histogram of words from the feature vectors.

Parallel attentive visual tracking

The research reported here addresses the problem of detecting and tracking independently moving objects from a moving observer in real-time, using corners as object tokens. Corners are detected using the Harris corner detector, and local image-plane constraints are employed to solve the correspondence problem. The approach relaxes the restrictive static-world assumption conventionally made, and is therefore capable of tracking independently moving and deformable objects. Tracking is performed without the use of any 3-dimensional motion model. The technique is novel in that, unlike traditional feature-tracking algorithms where feature detection and tracking is carried out over the entire image-plane, here it is restricted to those areas most likely to contain-meaningful image structure. Two distinct types of instantiation regions are identified, these being the “focus-of-expansion” region and “border” regions of the image-plane. The size and location of these regions are defined from a combination of odometry information and a limited knowledge of the operating scenario. The algorithms developed have been tested on real image sequences taken from typical driving scenarios. Implementation of the algorithm using T800 Transputers has shown that near-linear speedups are achievable, and that real-time operation is possible (half-video rate has been achieved using 30 processing elements).

Attentive visual tracking

The research reported here addresses the problem of detecting and tracking independently moving objects from a moving observer in real time, using corners as object tokens. Local image-plane constraints are employed to solve the correspondence problem removing the need for a 3D motion model. The approach relaxes the restrictive static-world assumption conventionally made, and is therefore capable of tracking independently moving and deformable objects. The technique is novel in that feature detection and tracking is restricted to areas likely to contain meaningful image structure. Feature instantiation regions are defined from a combination of odometry informatin and a limited knowledge of the operating scenario. The algorithms developed have been tested on real image sequences taken from typical driving scenarios. Preliminary experiments on a parallel (transputer) architecture indication that real-time operation is achievable.

Enhancing the teaching and learning of mathematical visual images

The literacy demands of mathematics are very different to those in other subjects (Gough, 2007; O'Halloran, 2005; Quinnell, 2011; Rubenstein, 2007) and much has been written on the challenges that literacy in mathematics poses to learners (Abedi and Lord, 2001; Lowrie and Diezmann, 2007, 2009; Rubenstein, 2007). In particular, a diverse selection of visuals typifies the field of mathematics (Carter, Hipwell and Quinnell, 2012), placing unique literacy demands on learners. Such visuals include varied tables, graphs, diagrams and other representations, all of which are used to communicate information.

From ImageNet to mining: Adapting visual object detection with minimal supervision

This paper presents visual detection and classification of light vehicles and personnel on a mine site.We capitalise on the rapid advances of ConvNet based object recognition but highlight that a naive black box approach results in a significant number of false positives. In particular, the lack of domain specific training data and the unique landscape in a mine site causes a high rate of errors. We exploit the abundance of background-only images to train a k-means classifier to complement the ConvNet. Furthermore, localisation of objects of interest and a reduction in computation is enabled through region proposals. Our system is tested on over 10km of real mine site data and we were able to detect both light vehicles and personnel. We show that the introduction of our background model can reduce the false positive rate by an order of magnitude.

The impact of simulated astigmatism on functional measures of visual performance

Uncorrected refractive error, including astigmatism, is a leading cause of reversible visual impairment. While the ability to perform vision-related daily activities is reduced when people are not optimally corrected, only limited research has investigated the impact of uncorrected astigmatism. Given the capacity to perform vision-related daily activities involves integration of a range of visual and cognitive cues, this research examined the impact of simulated astigmatism on visual tasks that also involved cognitive input. The research also examined whether the higher levels of complexity inherent in Chinese characters makes them more susceptible to the effects of astigmatism. The effects of different powers of astigmatism, as well as astigmatism at different axes were investigated in order to determine the minimum level of astigmatism that resulted in a decrement in visual performance.