Learning 3D structure from 2D images using LBP features

An automatic machine learning strategy for computing the 3D structure of monocular images from a single image query using Local Binary Patterns is presented. The 3D structure is inferred through a training set composed by a repository of color and depth images, assuming that images with similar structure present similar depth maps. Local Binary Patterns are used to characterize the structure of the color images. The depth maps of those color images with a similar structure to the query image are adaptively combined and filtered to estimate the final depth map. Using public databases, promising results have been obtained outperforming other state-of-the-art algorithms and with a computational cost similar to the most efficient 2D-to-3D algorithms.

Neural systems underlying learning and representation of global motion

We demonstrate performance-related changes in cortical and cerebellar activity. The largest learning-dependent changes were observed in the anterior lateral cerebellum, where the extent and intensity of activation correlated inversely with psychophysical performance. After learning had occurred (a few minutes), the cerebellar activation almost disappeared; however, it was restored when the subjects were presented with a novel, untrained direction of motion for which psychophysical performance also reverted to chance level. Similar reductions in the extent and intensity of brain activations in relation to learning occurred in the superior colliculus, anterior cingulate, and parts of the extrastriate cortex. The motion direction-sensitive middle temporal visual complex was a notable exception, where there was an expansion of the cortical territory activated by the trained stimulus. Together, these results indicate that the learning and representation of visual motion discrimination are mediated by different, but probably interacting, neuronal subsystems.

Perceptual learning in motion discrimination that generalizes across motion directions

When human subjects discriminate motion directions of two visual stimuli, their discrimination improves with practice. This improved performance has been found to be specific to the practiced directions and does not transfer to new motion directions. Indeed, such stimulus-specific learning has become a trademark finding in almost all perceptual learning studies and has been used to infer the loci of learning in the brain. For example, learning in motion discrimination has been inferred to occur in the visual area MT (medial temporal cortex) of primates, where neurons are selectively tuned to motion directions. However, such motion discrimination task is extremely difficult, as is typical of most perceptual learning tasks. When the difficulty is moderately reduced, learning transfers to new motion directions. This result challenges the idea of using simple visual stimuli to infer the locus of learning in low-level visual processes and suggests that higher-level processing is essential even in “simple” perceptual learning tasks.

Structure-activity relationships derived by machine learning: the use of atoms and their bond connectivities to predict mutagenicity by inductive logic programming.

We present a general approach to forming structure-activity relationships (SARs). This approach is based on representing chemical structure by atoms and their bond connectivities in combination with the inductive logic programming (ILP) algorithm PROGOL. Existing SAR methods describe chemical structure by using attributes which are general properties of an object. It is not possible to map chemical structure directly to attribute-based descriptions, as such descriptions have no internal organization. A more natural and general way to describe chemical structure is to use a relational description, where the internal construction of the description maps that of the object described. Our atom and bond connectivities representation is a relational description. ILP algorithms can form SARs with relational descriptions. We have tested the relational approach by investigating the SARs of 230 aromatic and heteroaromatic nitro compounds. These compounds had been split previously into two subsets, 188 compounds that were amenable to regression and 42 that were not. For the 188 compounds, a SAR was found that was as accurate as the best statistical or neural network-generated SARs. The PROGOL SAR has the advantages that it did not need the use of any indicator variables handcrafted by an expert, and the generated rules were easily comprehensible. For the 42 compounds, PROGOL formed a SAR that was significantly (P < 0.025) more accurate than linear regression, quadratic regression, and back-propagation. This SAR is based on an automatically generated structural alert for mutagenicity.

The wisdom of God manifested in the works of the creation : in two parts. Viz. The heavenly bodies, elements, meteors, fossils, vegetables, animals (beasts, birds, fishes, and insects), more particularly in the body of the earth, its figure, motion, and consistency, and in the admirable structure of the bodies of man and other animals, as also in their generation, &c. : with answers to some objections /

First published: 1691.

Calculating the Uncertainty of a Structure from Motion (SfM) Model, Cadman Quarry, Monroe, Washington

Structure from Motion (SfM) is a new form of photogrammetry that automates the rendering of georeferenced 3D models of objects using digital photographs and independently surveyed Ground Control Points (GCPs). This project seeks to quantify the error found in Digital Elevation Models (DEMs) produced using SfM. I modeled a rockslide found at the Cadman Quarry (Monroe, Washington) because the surface is vegetation-free, which is ideal for SfM and Terrestrial LiDAR Scanner (TLS) surveys. By using SfM, TLS, and GPS positioning at the same time, I attempted to find the deviation in the SfM model from the TLS model and GPS points. Using the deviation, I found the Root-Mean-Square Error (RMSE) between the SfM DEM and GPS positions. The RMSE of the SfM model when compared to surveyed GPS points is 17cm. I propagated the uncertainty of the GPS points with the RMSE of the SfM model to find the uncertainty of the SfM model compared to the NAD 1984 datum. The uncertainty of the SfM model compared to the NAD 1984 is 27cm. This study did not produce a model from the TLS that had sufficient resolution on horizontal surfaces to compare to surveyed GPS points.

Does industry matter? Examining the role of industry structure and organizational learning in innovation and brand performance

A manager's perception of industry structure (dynamism) has the potential to impact various organizational strategies and behaviors. This may be particularly so with regard to perceptions driving organizational learning orientations and innovation based marketing strategy. The position taken here suggests that firms operating within a competitive industry tend to pursue innovative ways of performing value-creating activities, which requires the development of learning capabilities. The results of a study of SMEs suggest that market focused learning, relative to other learning capabilities plays a key role in the relationships between industry structure, innovation and brand performance. The findings also show that market focused learning and internally focused learning influence innovation and that innovation influences a brand's performance. (c) 2005 Elsevier Inc. All rights reserved.

Supervised learning of graph structure

Graph-based representations have been used with considerable success in computer vision in the abstraction and recognition of object shape and scene structure. Despite this, the methodology available for learning structural representations from sets of training examples is relatively limited. In this paper we take a simple yet effective Bayesian approach to attributed graph learning. We present a naïve node-observation model, where we make the important assumption that the observation of each node and each edge is independent of the others, then we propose an EM-like approach to learn a mixture of these models and a Minimum Message Length criterion for components selection. Moreover, in order to avoid the bias that could arise with a single estimation of the node correspondences, we decide to estimate the sampling probability over all the possible matches. Finally we show the utility of the proposed approach on popular computer vision tasks such as 2D and 3D shape recognition. © 2011 Springer-Verlag.

Nonlinear soil-structure interaction for buried pressure pipes under differential ground motion

Pipelines extend thousands of kilometers across wide geographic areas as a network to provide essential services for modern life. It is inevitable that pipelines must pass through unfavorable ground conditions, which are susceptible to natural disasters. This thesis investigates the behaviour of buried pressure pipelines experiencing ground distortions induced by normal faulting. A recent large database of physical modelling observations on buried pipes of different stiffness relative to the surrounding soil subjected to normal faults provided a unique opportunity to calibrate numerical tools. Three-dimensional finite element models were developed to enable the complex soil-structure interaction phenomena to be further understood, especially on the subjects of gap formation beneath the pipe and the trench effect associated with the interaction between backfill and native soils. Benchmarked numerical tools were then used to perform parametric analysis regarding project geometry, backfill material, relative pipe-soil stiffness and pipe diameter. Seismic loading produces a soil displacement profile that can be expressed by isoil, the distance between the peak curvature and the point of contraflexure. A simplified design framework based on this length scale (i.e., the Kappa method) was developed, which features estimates of longitudinal bending moments of buried pipes using a characteristic length, ipipe, the distance from peak to zero curvature. Recent studies indicated that empirical soil springs that were calibrated against rigid pipes are not suitable for analyzing flexible pipes, since they lead to excessive conservatism (for design). A large-scale split-box normal fault simulator was therefore assembled to produce experimental data for flexible PVC pipe responses to a normal fault. Digital image correlation (DIC) was employed to analyze the soil displacement field, and both optical fibres and conventional strain gauges were used to measure pipe strains. A refinement to the Kappa method was introduced to enable the calculation of axial strains as a function of pipe elongation induced by flexure and an approximation of the longitudinal ground deformations. A closed-form Winkler solution of flexural response was also derived to account for the distributed normal fault pattern. Finally, these two analytical solutions were evaluated against the pipe responses observed in the large-scale laboratory tests.

Using UAV acquired photography and structure from motion techniques for studying glacier landforms: application to the glacial flutes at Isfallsglaciären

Glacier and ice sheet retreat exposes freshly deglaciated terrain which often contains small-scale fragile geomorphological features which could provide insight into subglacial or submarginal processes. Subaerial exposure results in potentially rapid landscape modification or even disappearance of the minor–relief landforms as wind, weather, water and vegetation impacts on the newly exposed surface. Ongoing retreat of many ice masses means there is a growing opportunity to obtain high resolution geospatial data from glacier forelands to aid in the understanding of recent subglacial and submarginal processes. Here we used an unmanned aerial vehicle to capture close-range aerial photography of the foreland of Isfallsglaciären, a small polythermal glacier situated in Swedish Lapland. An orthophoto and a digital elevation model with ~2 cm horizontal resolution were created from this photography using structure from motion software. These geospatial data was used to create a geomorphological map of the foreland, documenting moraines, fans, channels and flutes. The unprecedented resolution of the data enabled us to derive morphological metrics (length, width and relief) of the smallest flutes, which is not possible with other data products normally used for glacial landform metrics mapping. The map and flute metrics compare well with previous studies, highlighting the potential of this technique for rapidly documenting glacier foreland geomorphology at an unprecedented scale and resolution. The vast majority of flutes were found to have an associated stoss-side boulder, with the remainder having a likely explanation for boulder absence (burial or erosion). Furthermore, the size of this boulder was found to strongly correlate with the width and relief of the lee-side flute. This is consistent with the lee-side cavity infill model of flute formation. Whether this model is applicable to all flutes, or multiple mechanisms are required, awaits further study.

PIV measurements combined with the motion tracking technique to analyze flow around a moving porous structure

To gain a better understanding of the fluid–structure interaction and especially when dealing with a flow around an arbitrarily moving body, it is essential to develop measurement tools enabling the instantaneous detection of moving deformable interface during the flow measurements. A particularly useful application is the determination of unsteady turbulent flow velocity field around a moving porous fishing net structure which is of great interest for selectivity and also for the numerical code validation which needs a realistic database. To do this, a representative piece of fishing net structure is used to investigate both the Turbulent Boundary Layer (TBL) developing over the horizontal porous moving fishing net structure and the turbulent flow passing through the moving porous structure. For such an investigation, Time Resolved PIV measurements are carried out and combined with a motion tracking technique allowing the measurement of the instantaneous motion of the deformable fishing net during PIV measurements. Once the two-dimensional motion of the porous structure is accessed, PIV velocity measurements are analyzed in connection with the detected motion. Finally, the TBL is characterized and the effect of the structure motion on the volumetric flow rate passing though the moving porous structure is clearly demonstrated.

Spectral Frame Analysis and Learning through Graph Structure

This dissertation investigates the connection between spectral analysis and frame theory. When considering the spectral properties of a frame, we present a few novel results relating to the spectral decomposition. We first show that scalable frames have the property that the inner product of the scaling coefficients and the eigenvectors must equal the inverse eigenvalues. From this, we prove a similar result when an approximate scaling is obtained. We then focus on the optimization problems inherent to the scalable frames by first showing that there is an equivalence between scaling a frame and optimization problems with a non-restrictive objective function. Various objective functions are considered, and an analysis of the solution type is presented. For linear objectives, we can encourage sparse scalings, and with barrier objective functions, we force dense solutions. We further consider frames in high dimensions, and derive various solution techniques. From here, we restrict ourselves to various frame classes, to add more specificity to the results. Using frames generated from distributions allows for the placement of probabilistic bounds on scalability. For discrete distributions (Bernoulli and Rademacher), we bound the probability of encountering an ONB, and for continuous symmetric distributions (Uniform and Gaussian), we show that symmetry is retained in the transformed domain. We also prove several hyperplane-separation results. With the theory developed, we discuss graph applications of the scalability framework. We make a connection with graph conditioning, and show the in-feasibility of the problem in the general case. After a modification, we show that any complete graph can be conditioned. We then present a modification of standard PCA (robust PCA) developed by Cand\`es, and give some background into Electron Energy-Loss Spectroscopy (EELS). We design a novel scheme for the processing of EELS through robust PCA and least-squares regression, and test this scheme on biological samples. Finally, we take the idea of robust PCA and apply the technique of kernel PCA to perform robust manifold learning. We derive the problem and present an algorithm for its solution. There is also discussion of the differences with RPCA that make theoretical guarantees difficult.