A method to quantify a descriptor's illumination variance

This paper presents a new metric, which we call the lighting variance ratio, for quantifying descriptors in terms of their variance to illumination changes. In many applications it is desirable to have descriptors that are robust to changes in illumination, especially in outdoor environments. The lighting variance ratio is useful for comparing descriptors and determining if a descriptor is lighting invariant enough for a given environment. The metric is analysed across a number of datasets, cameras and descriptors. The results show that the upright SIFT descriptor is typically the most lighting invariant descriptor.

Robotics for the mining industry

The mining industry is highly suitable for the application of robotics and automation technology since the work is both arduous and dangerous. However, while the industry makes extensive use of mechanisation it has shown a slow uptake of automation. A major cause of this is the complexity of the task, and the limitations of existing automation technology which is predicated on a structured and time invariant working environment. Here we discuss the topic of mining automation from a robotics and computer vision perspective — as a problem in sensor based robot control, an issue which the robotics community has been studying for nearly two decades. We then describe two of our current mining automation projects to demonstrate what is possible for both open-pit and underground mining operations.

Automatic image scaling for place recognition in changing environments

Robustness to variations in environmental conditions and camera viewpoint is essential for long-term place recognition, navigation and SLAM. Existing systems typically solve either of these problems, but invariance to both remains a challenge. This paper presents a training-free approach to lateral viewpoint- and condition-invariant, vision-based place recognition. Our successive frame patch-tracking technique infers average scene depth along traverses and automatically rescales views of the same place at different depths to increase their similarity. We combine our system with the condition-invariant SMART algorithm and demonstrate place recognition between day and night, across entire 4-lane-plus-median-strip roads, where current algorithms fail.

Brain fiber architecture, genetics, and intelligence: a high angular resolution diffusion imaging (HARDI) study

We developed an analysis pipeline enabling population studies of HARDI data, and applied it to map genetic influences on fiber architecture in 90 twin subjects. We applied tensor-driven 3D fluid registration to HARDI, resampling the spherical fiber orientation distribution functions (ODFs) in appropriate Riemannian manifolds, after ODF regularization and sharpening. Fitting structural equation models (SEM) from quantitative genetics, we evaluated genetic influences on the Jensen-Shannon divergence (JSD), a novel measure of fiber spatial coherence, and on the generalized fiber anisotropy (GFA) a measure of fiber integrity. With random-effects regression, we mapped regions where diffusion profiles were highly correlated with subjects' intelligence quotient (IQ). Fiber complexity was predominantly under genetic control, and higher in more highly anisotropic regions; the proportion of genetic versus environmental control varied spatially. Our methods show promise for discovering genes affecting fiber connectivity in the brain.

Direct mapping of hippocampal surfaces with intrinsic shape context

We propose in this paper a new method for the mapping of hippocampal (HC) surfaces to establish correspondences between points on HC surfaces and enable localized HC shape analysis. A novel geometric feature, the intrinsic shape context, is defined to capture the global characteristics of the HC shapes. Based on this intrinsic feature, an automatic algorithm is developed to detect a set of landmark curves that are stable across population. The direct map between a source and target HC surface is then solved as the minimizer of a harmonic energy function defined on the source surface with landmark constraints. For numerical solutions, we compute the map with the approach of solving partial differential equations on implicit surfaces. The direct mapping method has the following properties: (1) it has the advantage of being automatic; (2) it is invariant to the pose of HC shapes. In our experiments, we apply the direct mapping method to study temporal changes of HC asymmetry in Alzheimer's disease (AD) using HC surfaces from 12 AD patients and 14 normal controls. Our results show that the AD group has a different trend in temporal changes of HC asymmetry than the group of normal controls. We also demonstrate the flexibility of the direct mapping method by applying it to construct spherical maps of HC surfaces. Spherical harmonics (SPHARM) analysis is then applied and it confirms our results on temporal changes of HC asymmetry in AD.

Increasing functional variability in the preparatory phase of the takeoff improves elite springboard diving performance

PURPOSE: Previous research demonstrating that specific performance outcome goals can be achieved in different ways is functionally significant for springboard divers whose performance environment can vary extensively. This body of work raises questions about the traditional approach of balking (terminating the takeoff) by elite divers aiming to perform only identical, invariant movement patterns during practice. METHOD: A 12-week training program (2 times per day; 6.5 hr per day) was implemented with 4 elite female springboard divers to encourage them to adapt movement patterns under variable takeoff conditions and complete intended dives, rather than balk. RESULTS: Intraindividual analyses revealed small increases in variability in the board-work component of each diver's pretraining and posttraining program reverse-dive takeoffs. No topological differences were observed between movement patterns of dives completed pretraining and posttraining. Differences were noted in the amount of movement variability under different training conditions (evidenced by higher normalized root mean square error indexes posttraining). An increase in the number of completed dives (from 78.91%-86.84% to 95.59%-99.29%) and a decrease in the frequency of balked takeoffs (from 13.16%-19.41% to 0.63%-4.41%) showed that the elite athletes were able to adapt their behaviors during the training program. These findings coincided with greater consistency in the divers' performance during practice as scored by qualified judges. CONCLUSION: Results suggested that on completion of training, athletes were capable of successfully adapting their movement patterns under more varied takeoff conditions to achieve greater consistency and stability of performance outcomes.

MEG Adaptation resolves the spatiotemporal characteristics of face-sensitive brain responses

An unresolved goal in face perception is to identify brain areas involved in face processing and simultaneously understand the timing of their involvement. Currently, high spatial resolution imaging techniques identify the fusiform gyrus as subserving processing of invariant face features relating to identity. High temporal resolution imaging techniques localize an early latency evoked component—the N/M170—as having a major generator in the fusiform region; however, this evoked component is not believed to be associated with the processing of identity. To resolve this, we used novel magnetoencephalographic beamformer analyses to localize cortical regions in humans spatially with trial-by-trial activity that differentiated faces and objects and to interrogate their functional sensitivity by analyzing the effects of stimulus repetition. This demonstrated a temporal sequence of processing that provides category-level and then item-level invariance. The right fusiform gyrus showed adaptation to faces (not objects) at ∼150 ms after stimulus onset regardless of face identity; however, at the later latency of ∼200–300 ms, this area showed greater adaptation to repeated identity faces than to novel identities. This is consistent with an involvement of the fusiform region in both early and midlatency face-processing operations, with only the latter showing sensitivity to invariant face features relating to identity.

MRI-based strain and strain rate analysis of left ventricle: A modified hierarchical transformation model

Background Different from other indicators of cardiac function, such as ejection fraction and transmitral early diastolic velocity, myocardial strain is promising to capture subtle alterations that result from early diseases of the myocardium. In order to extract the left ventricle (LV) myocardial strain and strain rate from cardiac cine-MRI, a modified hierarchical transformation model was proposed. Methods A hierarchical transformation model including the global and local LV deformations was employed to analyze the strain and strain rate of the left ventricle by cine-MRI image registration. The endocardial and epicardial contour information was introduced to enhance the registration accuracy by combining the original hierarchical algorithm with an Iterative Closest Points using Invariant Features algorithm. The hierarchical model was validated by a normal volunteer first and then applied to two clinical cases (i.e., the normal volunteer and a diabetic patient) to evaluate their respective function. Results Based on the two clinical cases, by comparing the displacement fields of two selected landmarks in the normal volunteer, the proposed method showed a better performance than the original or unmodified model. Meanwhile, the comparison of the radial strain between the volunteer and patient demonstrated their apparent functional difference. Conclusions The present method could be used to estimate the LV myocardial strain and strain rate during a cardiac cycle and thus to quantify the analysis of the LV motion function.

Smooth stabilisation of nonholonomic robots subject to disturbances

In this paper, we address the problem of stabilisation of robots subject to nonholonommic constraints and external disturbances using port-Hamiltonian theory and smooth time-invariant control laws. This should be contrasted with the commonly used switched or time-varying laws. We propose a control design that provides asymptotic stability of an manifold (also called relative equilibria)-due to the Brockett condition this is the only type of stabilisation possible using smooth time-invariant control laws. The equilibrium manifold can be shaped to certain extent to satisfy specific control objectives. The proposed control law also incorporates integral action, and thus the closed-loop system is robust to unknown constant disturbances. A key step in the proposed design is a change of coordinates not only in the momentum, but also in the position vector, which differs from coordinate transformations previously proposed in the literature for the control of nonholonomic systems. The theoretical properties of the control law are verified via numerical simulation based on a robotic ground vehicle model with differential traction wheels and non co-axial centre of mass and point of contact.

Overfitting Bayesian mixture models with an unknown number of components

This paper proposes solutions to three issues pertaining to the estimation of finite mixture models with an unknown number of components: the non-identifiability induced by overfitting the number of components, the mixing limitations of standard Markov Chain Monte Carlo (MCMC) sampling techniques, and the related label switching problem. An overfitting approach is used to estimate the number of components in a finite mixture model via a Zmix algorithm. Zmix provides a bridge between multidimensional samplers and test based estimation methods, whereby priors are chosen to encourage extra groups to have weights approaching zero. MCMC sampling is made possible by the implementation of prior parallel tempering, an extension of parallel tempering. Zmix can accurately estimate the number of components, posterior parameter estimates and allocation probabilities given a sufficiently large sample size. The results will reflect uncertainty in the final model and will report the range of possible candidate models and their respective estimated probabilities from a single run. Label switching is resolved with a computationally light-weight method, Zswitch, developed for overfitted mixtures by exploiting the intuitiveness of allocation-based relabelling algorithms and the precision of label-invariant loss functions. Four simulation studies are included to illustrate Zmix and Zswitch, as well as three case studies from the literature. All methods are available as part of the R package Zmix, which can currently be applied to univariate Gaussian mixture models.

Domain Adaptation on the Statistical Manifold

In this paper, we tackle the problem of unsupervised domain adaptation for classification. In the unsupervised scenario where no labeled samples from the target domain are provided, a popular approach consists in transforming the data such that the source and target distributions be- come similar. To compare the two distributions, existing approaches make use of the Maximum Mean Discrepancy (MMD). However, this does not exploit the fact that prob- ability distributions lie on a Riemannian manifold. Here, we propose to make better use of the structure of this man- ifold and rely on the distance on the manifold to compare the source and target distributions. In this framework, we introduce a sample selection method and a subspace-based method for unsupervised domain adaptation, and show that both these manifold-based techniques outperform the cor- responding approaches based on the MMD. Furthermore, we show that our subspace-based approach yields state-of- the-art results on a standard object recognition benchmark.

NKT cells from normal and tumor-bearing human livers are phenotypically and functionally distinct from murine NKT cells

A major group of murine NK T (NKT) cells express an invariant Vα14Jα18 TCR α-chain specific for glycolipid Ags presented by CD1d. Murine Vα14Jα18+ account for 30–50% of hepatic T cells and have potent antitumor activities. We have enumerated and characterized their human counterparts, Vα24Vβ11+ NKT cells, freshly isolated from histologically normal and tumor-bearing livers. In contrast to mice, human NKT cells are found in small numbers in healthy liver (0.5% of CD3+ cells) and blood (0.02%). In contrast to those in blood, most hepatic Vα24+ NKT cells express the Vβ11 chain. They include CD4+, CD8+, and CD4−CD8− cells, and many express the NK cell markers CD56, CD161, and/or CD69. Importantly, human hepatic Vα24+ T cells are potent producers of IFN-γ and TNF-α, but not IL-2 or IL-4, when stimulated pharmacologically or with the NKT cell ligand, α-galactosylceramide. Vα24+Vβ11+ cell numbers are reduced in tumor-bearing compared with healthy liver (0.1 vs 0.5%; p < 0.04). However, hepatic cells from cancer patients and healthy donors release similar amounts of IFN-γ in response to α-galactosylceramide. These data indicate that hepatic NKT cell repertoires are phenotypically and functionally distinct in humans and mice. Depletions of hepatic NKT cell subpopulations may underlie the susceptibility to metastatic liver disease.

CD1 expression and CD1-restricted T cell activity in normal and tumour-bearing human liver

CD1d-restricted natural killer T (NKT) cells expressing invariant Valpha14Jalpha18 T cell receptor alpha-chains are abundant in murine liver and are implicated in the control of malignancy, infection and autoimmunity. Invariant NKT cells have potent anti-metastatic effects in mice and phase I clinical trials involving their homologues in humans are ongoing. However, invariant NKT cells are less abundant in human liver ( approximately 0.5% of hepatic T cells) than in murine liver (up to 50%) and it is not known if other hepatic T cells are CD1-restricted. We have examined expression of CD1a, CD1b, CD1c and CD1d mRNA and protein in human liver and evaluated the reactivity of mononuclear cells (MNC) from histologically normal and tumour-bearing human liver specimens against these CD1 isoforms. Messenger RNA for all CD1 isotypes was detectable in all liver samples. CD1c and CD1d were expressed at the protein level by hepatic MNC. CD1d, only, was detectable at the cell surface, but CD1c and CD1d were found at an intracellular location in significant numbers of liver MNC. CD1b was not expressed by MNC from healthy livers but was detectable within MNC in all tumour samples tested. Hepatic T cells exhibited reactivity against C1R cells expressing transfected CD1c and CD1d, but neither CD1a nor CD1b. These cells secreted interferon-gamma (IFN-gamma) but not interleukin-4 (IL-4) upon stimulation. In contrast, similar numbers of peripheral T cells released 13- and 16-fold less IFN-gamma in response to CD1c and CD1d, respectively. CD1c and CD1d expression and T cell reactivity were not altered in tumour-bearing liver specimens compared to histologically normal livers. These data suggest that, in addition to invariant CD1d-restricted NKT cells, autoreactive T cells that recognise CD1c and CD1d and release inflammatory cytokines are abundant in human liver.

Stationary processes for object detection and non-rigid structure-from-motion

Stationary processes are random variables whose value is a signal and whose distribution is invariant to translation in the domain of the signal. They are intimately connected to convolution, and therefore to the Fourier transform, since the covariance matrix of a stationary process is a Toeplitz matrix, and Toeplitz matrices are the expression of convolution as a linear operator. This thesis utilises this connection in the study of i) efficient training algorithms for object detection and ii) trajectory-based non-rigid structure-from-motion.