Parallel Computer Vision Algorithms for Graphics Processing Units

La evolución de los teléfonos móviles inteligentes, dotados de cámaras digitales, está provocando una creciente demanda de aplicaciones cada vez más complejas que necesitan algoritmos de visión artificial en tiempo real; puesto que el tamaño de las señales de vídeo no hace sino aumentar y en cambio el rendimiento de los procesadores de un solo núcleo se ha estancado, los nuevos algoritmos que se diseñen para visión artificial han de ser paralelos para poder ejecutarse en múltiples procesadores y ser computacionalmente escalables. Una de las clases de procesadores más interesantes en la actualidad se encuentra en las tarjetas gráficas (GPU), que son dispositivos que ofrecen un alto grado de paralelismo, un excelente rendimiento numérico y una creciente versatilidad, lo que los hace interesantes para llevar a cabo computación científica. En esta tesis se exploran dos aplicaciones de visión artificial que revisten una gran complejidad computacional y no pueden ser ejecutadas en tiempo real empleando procesadores tradicionales. En cambio, como se demuestra en esta tesis, la paralelización de las distintas subtareas y su implementación sobre una GPU arrojan los resultados deseados de ejecución con tasas de refresco interactivas. Asimismo, se propone una técnica para la evaluación rápida de funciones de complejidad arbitraria especialmente indicada para su uso en una GPU. En primer lugar se estudia la aplicación de técnicas de síntesis de imágenes virtuales a partir de únicamente dos cámaras lejanas y no paralelas—en contraste con la configuración habitual en TV 3D de cámaras cercanas y paralelas—con información de color y profundidad. Empleando filtros de mediana modificados para la elaboración de un mapa de profundidad virtual y proyecciones inversas, se comprueba que estas técnicas son adecuadas para una libre elección del punto de vista. Además, se demuestra que la codificación de la información de profundidad con respecto a un sistema de referencia global es sumamente perjudicial y debería ser evitada. Por otro lado se propone un sistema de detección de objetos móviles basado en técnicas de estimación de densidad con funciones locales. Este tipo de técnicas es muy adecuada para el modelado de escenas complejas con fondos multimodales, pero ha recibido poco uso debido a su gran complejidad computacional. El sistema propuesto, implementado en tiempo real sobre una GPU, incluye propuestas para la estimación dinámica de los anchos de banda de las funciones locales, actualización selectiva del modelo de fondo, actualización de la posición de las muestras de referencia del modelo de primer plano empleando un filtro de partículas multirregión y selección automática de regiones de interés para reducir el coste computacional. Los resultados, evaluados sobre diversas bases de datos y comparados con otros algoritmos del estado del arte, demuestran la gran versatilidad y calidad de la propuesta. Finalmente se propone un método para la aproximación de funciones arbitrarias empleando funciones continuas lineales a tramos, especialmente indicada para su implementación en una GPU mediante el uso de las unidades de filtraje de texturas, normalmente no utilizadas para cómputo numérico. La propuesta incluye un riguroso análisis matemático del error cometido en la aproximación en función del número de muestras empleadas, así como un método para la obtención de una partición cuasióptima del dominio de la función para minimizar el error. ABSTRACT The evolution of smartphones, all equipped with digital cameras, is driving a growing demand for ever more complex applications that need to rely on real-time computer vision algorithms. However, video signals are only increasing in size, whereas the performance of single-core processors has somewhat stagnated in the past few years. Consequently, new computer vision algorithms will need to be parallel to run on multiple processors and be computationally scalable. One of the most promising classes of processors nowadays can be found in graphics processing units (GPU). These are devices offering a high parallelism degree, excellent numerical performance and increasing versatility, which makes them interesting to run scientific computations. In this thesis, we explore two computer vision applications with a high computational complexity that precludes them from running in real time on traditional uniprocessors. However, we show that by parallelizing subtasks and implementing them on a GPU, both applications attain their goals of running at interactive frame rates. In addition, we propose a technique for fast evaluation of arbitrarily complex functions, specially designed for GPU implementation. First, we explore the application of depth-image–based rendering techniques to the unusual configuration of two convergent, wide baseline cameras, in contrast to the usual configuration used in 3D TV, which are narrow baseline, parallel cameras. By using a backward mapping approach with a depth inpainting scheme based on median filters, we show that these techniques are adequate for free viewpoint video applications. In addition, we show that referring depth information to a global reference system is ill-advised and should be avoided. Then, we propose a background subtraction system based on kernel density estimation techniques. These techniques are very adequate for modelling complex scenes featuring multimodal backgrounds, but have not been so popular due to their huge computational and memory complexity. The proposed system, implemented in real time on a GPU, features novel proposals for dynamic kernel bandwidth estimation for the background model, selective update of the background model, update of the position of reference samples of the foreground model using a multi-region particle filter, and automatic selection of regions of interest to reduce computational cost. The results, evaluated on several databases and compared to other state-of-the-art algorithms, demonstrate the high quality and versatility of our proposal. Finally, we propose a general method for the approximation of arbitrarily complex functions using continuous piecewise linear functions, specially formulated for GPU implementation by leveraging their texture filtering units, normally unused for numerical computation. Our proposal features a rigorous mathematical analysis of the approximation error in function of the number of samples, as well as a method to obtain a suboptimal partition of the domain of the function to minimize approximation error.

Grid-Connected Forward Microinverter With Primary-Parallel Secondary-Series Transformer

This paper presents a primary-parallel secondaryseries multicore forward microinverter for photovoltaic ac-module application. The presented microinverter operates with a constant off-time boundary mode control, providing MPPT capability and unity power factor. The proposed multitransformer solution allows using low-profile unitary turns ratio transformers. Therefore, the transformers are better coupled and the overall performance of the microinverter is improved. Due to the multiphase solution, the number of devices increases but the current stress and losses per device are reduced contributing to an easier thermal management. Furthermore, the decoupling capacitor is split among the phases, contributing to a low-profile solution without electrolytic capacitors suitable to be mounted in the frame of a PV module. The proposed solution is compared to the classical parallel-interleaved approach, showing better efficiency in a wide power range and improving the weighted efficiency.

Parallel-up structure evidences the molecular directionality during biosynthesis of bacterial cellulose

The “parallel-up” packing in cellulose Iα and Iβ unit cells was experimentally demonstrated by a combination of direct-staining the reducing ends of cellulose chains and microdiffraction-tilting electron crystallographic analysis. Microdiffraction investigation of nascent bacterial cellulose microfibrils showed that the reducing end of the growing cellulose chains points away from the bacterium, and this provides direct evidence that polymerization by the cellulose synthase takes place at the nonreducing end of the growing cellulose chains. This mechanism is likely to be valid also for a number of processive glycosyltransferases such as chitin synthases, hyaluronan synthases, and proteins involved in the synthesis of nodulation factor backbones.

Yeast microarrays for genome wide parallel genetic and gene expression analysis

We have developed high-density DNA microarrays of yeast ORFs. These microarrays can monitor hybridization to ORFs for applications such as quantitative differential gene expression analysis and screening for sequence polymorphisms. Automated scripts retrieved sequence information from public databases to locate predicted ORFs and select appropriate primers for amplification. The primers were used to amplify yeast ORFs in 96-well plates, and the resulting products were arrayed using an automated micro arraying device. Arrays containing up to 2,479 yeast ORFs were printed on a single slide. The hybridization of fluorescently labeled samples to the array were detected and quantitated with a laser confocal scanning microscope. Applications of the microarrays are shown for genetic and gene expression analysis at the whole genome level.

Patches of synchronized activity in the cerebellar cortex evoked by mossy-fiber stimulation: Questioning the role of parallel fibers

The discrepancy between the structural longitudinal organization of the parallel-fiber system in the cerebellar cortex and the functional mosaic-like organization of the cortex has provoked controversial theories about the flow of information in the cerebellum. We address this issue by characterizing the spatiotemporal organization of neuronal activity in the cerebellar cortex by using optical imaging of voltage-sensitive dyes in isolated guinea-pig cerebellum. Parallel-fiber stimulation evoked a narrow beam of activity, which propagated along the parallel fibers. Stimulation of the mossy fibers elicited a circular, nonpropagating patch of synchronized activity. These results strongly support the hypothesis that a beam of parallel fibers, activated by a focal group of granule cells, fails to activate the Purkinje cells along most of its length. It is thus the ascending axon of the granule cell, and not its parallel branches, that activates and defines the basic functional modules of the cerebellar cortex.

Propagating structure of Alzheimer’s β-amyloid(10–35) is parallel β-sheet with residues in exact register

The pathognomonic plaques of Alzheimer’s disease are composed primarily of the 39- to 43-aa β-amyloid (Aβ) peptide. Crosslinking of Aβ peptides by tissue transglutaminase (tTg) indicates that Gln15 of one peptide is proximate to Lys16 of another in aggregated Aβ. Here we report how the fibril structure is resolved by mapping interstrand distances in this core region of the Aβ peptide chain with solid-state NMR. Isotopic substitution provides the source points for measuring distances in aggregated Aβ. Peptides containing a single carbonyl 13C label at Gln15, Lys16, Leu17, or Val18 were synthesized and evaluated by NMR dipolar recoupling methods for the measurement of interpeptide distances to a resolution of 0.2 Å. Analysis of these data establish that this central core of Aβ consists of a parallel β-sheet structure in which identical residues on adjacent chains are aligned directly, i.e., in register. Our data, in conjunction with existing structural data, establish that the Aβ fibril is a hydrogen-bonded, parallel β-sheet defining the long axis of the Aβ fibril propagation.

Transformation by v-Src: Ras-MAPK and PI3K-mTOR Mediate Parallel Pathways

An increase in the level of active, GTP-bound Ras is not necessary for transformation of chicken embryo fibroblasts (CEF) by v-Src. This suggests that other Ras-independent pathways contribute to transformation by v-Src. To address the possibility that activation of phosphatidylinositol-3-kinase (PI3K) and the mammalian target of rapamycin (mTOR/FRAP), represents one of these pathways, we have examined the effect of simultaneous inhibition of the Ras-MAPK and PI3K-mTOR pathways on transformation of CEF by v-Src. Transformation was assessed by the standard parameters of morphological alteration, increased hexose uptake, loss of density inhibition, and anchorage-independent growth. Inhibition of the Ras-MAPK pathway by expression of the dominant-negative Ras mutant HRasN17 or by addition of the MAPK kinase (MEK) inhibitor PD98059 reduced several of these parameters but failed to block transformation. Similarly, inhibition of the PI3K-mTOR pathway by addition of the PI3K inhibitor 2-[4-morpholinyl]-8-phenyl-4H-1-benzopyran-4-one (LY294002) or the mTOR inhibitor rapamycin, although reducing several parameters of transformation, also failed to block transformation. However, simultaneous inhibition of signaling by the Ras-MAPK pathway and the PI3K-mTOR pathway essentially blocked transformation. These data indicate that transformation of CEF by v-Src is mediated by two parallel pathways, the Ras-MAPK pathway and the PI-3K-mTOR pathway, which both contribute to transformation. The possibility that simultaneous activation of other pathways is also required is not excluded.

Automated parallel DNA sequencing on multiple channel microchips

We report automated DNA sequencing in 16-channel microchips. A microchip prefilled with sieving matrix is aligned on a heating plate affixed to a movable platform. Samples are loaded into sample reservoirs by using an eight-tip pipetting device, and the chip is docked with an array of electrodes in the focal plane of a four-color scanning detection system. Under computer control, high voltage is applied to the appropriate reservoirs in a programmed sequence that injects and separates the DNA samples. An integrated four-color confocal fluorescent detector automatically scans all 16 channels. The system routinely yields more than 450 bases in 15 min in all 16 channels. In the best case using an automated base-calling program, 543 bases have been called at an accuracy of >99%. Separations, including automated chip loading and sample injection, normally are completed in less than 18 min. The advantages of DNA sequencing on capillary electrophoresis chips include uniform signal intensity and tolerance of high DNA template concentration. To understand the fundamentals of these unique features we developed a theoretical treatment of cross-channel chip injection that we call the differential concentration effect. We present experimental evidence consistent with the predictions of the theory.

Estrogen-induced transcription of the progesterone receptor gene does not parallel estrogen receptor occupancy

The activation of the silent endogenous progesterone receptor (PR) gene by 17-β-estradiol (E2) in cells stably transfected with estrogen receptor (ER) was used as a model system to study the mechanism of E2-induced transcription. The time course of E2-induced PR transcription rate was determined by nuclear run-on assays. No marked effect on specific PR gene transcription rates was detected at 0 and 1 h of E2 treatment. After 3 h of E2 treatment, the PR mRNA synthesis rate increased 2.0- ± 0.2-fold and continued to increase to 3.5- ± 0.4-fold by 24 h as compared with 0 h. The transcription rate increase was followed by PR mRNA accumulation. No PR mRNA was detectable at 0, 1, and 3 h of E2 treatment. PR mRNA accumulation was detected at 6 h of E2 treatment and continued to accumulate until 18 h, the longest time point examined. Interestingly, this slow and gradual transcription rate increase of the endogenous PR gene did not parallel binding of E2 to ER, which was maximized within 30 min. Furthermore, the E2–ER level was down-regulated to 15% at 3 h as compared with 30 min of E2 treatment and remained low at 24 h of E2 exposure. These paradoxical observations indicate that E2-induced transcription activation is more complicated than just an association of the occupied ER with the transcription machinery.

Multiple quantum solid-state NMR indicates a parallel, not antiparallel, organization of β-sheets in Alzheimer's β-amyloid fibrils

Senile plaques associated with Alzheimer's disease contain deposits of fibrils formed by 39- to 43-residue β-amyloid peptides with possible neurotoxic effects. X-ray diffraction measurements on oriented fibril bundles have indicated an extended β-sheet structure for Alzheimer's β-amyloid fibrils and other amyloid fibrils, but the supramolecular organization of the β-sheets and other structural details are not well established because of the intrinsically noncrystalline, insoluble nature of amyloid fibrils. Here we report solid-state NMR measurements, using a multiple quantum (MQ) 13C NMR technique, that probe the β-sheet organization in fibrils formed by the full-length, 40-residue β-amyloid peptide (Aβ1–40). Although an antiparallel β-sheet organization often is assumed and is invoked in recent structural models for full-length β-amyloid fibrils, the MQNMR data indicate an in-register, parallel organization. This work provides site-specific, atomic-level structural constraints on full-length β-amyloid fibrils and applies MQNMR to a significant problem in structural biology.

A model of long-term memory storage in the cerebellar cortex: A possible role for plasticity at parallel fiber synapses onto stellate/basket interneurons

By evoking changes in climbing fiber activity, movement errors are thought to modify synapses from parallel fibers onto Purkinje cells (pf*Pkj) so as to improve subsequent motor performance. Theoretical arguments suggest there is an intrinsic tradeoff, however, between motor adaptation and long-term storage. Assuming a baseline rate of motor errors is always present, then repeated performance of any learned movement will generate a series of climbing fiber-mediated corrections. By reshuffling the synaptic weights responsible for any given movement, such corrections will degrade the memories for other learned movements stored in overlapping sets of synapses. The present paper shows that long-term storage can be accomplished by a second site of plasticity at synapses from parallel fibers onto stellate/basket interneurons (pf*St/Bk). Plasticity at pf*St/Bk synapses can be insulated from ongoing fluctuations in climbing fiber activity by assuming that changes in pf*St/Bk synapses occur only after changes in pf*Pkj synapses have built up to a threshold level. Although climbing fiber-dependent plasticity at pf*Pkj synapses allows for the exploration of novel motor strategies in response to changing environmental conditions, plasticity at pf*St/Bk synapses transfers successful strategies to stable long-term storage. To quantify this hypothesis, both sites of plasticity are incorporated into a dynamical model of the cerebellar cortex and its interactions with the inferior olive. When used to simulate idealized motor conditioning trials, the model predicts that plasticity develops first at pf*Pkj synapses, but with additional training is transferred to pf*St/Bk synapses for long-term storage.

Protein-free parallel triple-stranded DNA complex formation

A 14 nt DNA sequence 5′-AGAATGTGGCAAAG-3′ from the zinc finger repeat of the human KRAB zinc finger protein gene ZNF91 bearing the intercalator 2-methoxy,6-chloro,9-amino acridine (Acr) attached to the sugar–phosphate backbone in various positions has been shown to form a specific triple helix (triplex) with a 16 bp hairpin (intramolecular) or a two-stranded (intermolecular) duplex having the identical sequence in the same (parallel) orientation. Intramolecular targets with the identical sequence in the antiparallel orientation and a non-specific target sequence were tested as controls. Apparent binding constants for formation of the triplex were determined by quantitating electrophoretic band shifts. Binding of the single-stranded oligonucleotide probe sequence to the target led to an increase in the fluorescence anisotropy of acridine. The parallel orientation of the two identical sequence segments was confirmed by measurement of fluorescence resonance energy transfer between the acridine on the 5′-end of the probe strand as donor and BODIPY-Texas Red on the 3′-amino group of either strand of the target duplex as acceptor. There was full protection from OsO4-bipyridine modification of thymines in the probe strand of the triplex, in accordance with the presumed triplex formation, which excluded displacement of the homologous duplex strand by the probe–intercalator conjugate. The implications of these results for the existence of protein-independent parallel triplexes are discussed.

ParAlign: a parallel sequence alignment algorithm for rapid and sensitive database searches

There is a need for faster and more sensitive algorithms for sequence similarity searching in view of the rapidly increasing amounts of genomic sequence data available. Parallel processing capabilities in the form of the single instruction, multiple data (SIMD) technology are now available in common microprocessors and enable a single microprocessor to perform many operations in parallel. The ParAlign algorithm has been specifically designed to take advantage of this technology. The new algorithm initially exploits parallelism to perform a very rapid computation of the exact optimal ungapped alignment score for all diagonals in the alignment matrix. Then, a novel heuristic is employed to compute an approximate score of a gapped alignment by combining the scores of several diagonals. This approximate score is used to select the most interesting database sequences for a subsequent Smith–Waterman alignment, which is also parallelised. The resulting method represents a substantial improvement compared to existing heuristics. The sensitivity and specificity of ParAlign was found to be as good as Smith–Waterman implementations when the same method for computing the statistical significance of the matches was used. In terms of speed, only the significantly less sensitive NCBI BLAST 2 program was found to outperform the new approach. Online searches are available at http://dna.uio.no/search/

The CONSORT statement: revised recommendations for improving the quality of reports of parallel group randomized trials

To comprehend the results of a randomized controlled trial (RCT), readers must understand its design, conduct, analysis and interpretation. That goal can only be achieved through complete transparency from authors. Despite several decades of educational efforts, the reporting of RCTs needs improvement. Investigators and editors developed the original CONSORT (Consolidated Standards of Reporting Trials) statement to help authors improve reporting by using a checklist and flow diagram. The revised CONSORT statement presented in this paper incorporates new evidence and addresses some criticisms of the original statement.