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.

Particle methods parallel implementations by GP-GPU strategies

This paper outlines the problems found in the parallelization of SPH (Smoothed Particle Hydrodynamics) algorithms using Graphics Processing Units. Different results of some parallel GPU implementations in terms of the speed-up and the scalability compared to the CPU sequential codes are shown. The most problematic stage in the GPU-SPH algorithms is the one responsible for locating neighboring particles and building the vectors where this information is stored, since these specific algorithms raise many dificulties for a data-level parallelization. Because of the fact that the neighbor location using linked lists does not show enough data-level parallelism, two new approaches have been pro- posed to minimize bank conflicts in the writing and subsequent reading of the neighbor lists. The first strategy proposes an efficient coordination between CPU-GPU, using GPU algorithms for those stages that allow a straight forward parallelization, and sequential CPU algorithms for those instructions that involve some kind of vector reduction. This coordination provides a relatively orderly reading of the neighbor lists in the interactions stage, achieving a speed-up factor of x47 in this stage. However, since the construction of the neighbor lists is quite expensive, it is achieved an overall speed-up of x41. The second strategy seeks to maximize the use of the GPU in the neighbor's location process by executing a specific vector sorting algorithm that allows some data-level parallelism. Al- though this strategy has succeeded in improving the speed-up on the stage of neighboring location, the global speed-up on the interactions stage falls, due to inefficient reading of the neighbor vectors. Some changes to these strategies are proposed, aimed at maximizing the computational load of the GPU and using the GPU texture-units, in order to reach the maximum speed-up for such codes. Different practical applications have been added to the mentioned GPU codes. First, the classical dam-break problem is studied. Second, the wave impact of the sloshing fluid contained in LNG vessel tanks is also simulated as a practical example of particle methods

A Novel Scalable Deblocking-Filter Architecture for H.264/AVC and SVC Video Codecs

A highly parallel and scalable Deblocking Filter (DF) hardware architecture for H.264/AVC and SVC video codecs is presented in this paper. The proposed architecture mainly consists on a coarse grain systolic array obtained by replicating a unique and homogeneous Functional Unit (FU), in which a whole Deblocking-Filter unit is implemented. The proposal is also based on a novel macroblock-level parallelization strategy of the filtering algorithm which improves the final performance by exploiting specific data dependences. This way communication overhead is reduced and a more intensive parallelism in comparison with the existing state-of-the-art solutions is obtained. Furthermore, the architecture is completely flexible, since the level of parallelism can be changed, according to the application requirements. The design has been implemented in a Virtex-5 FPGA, and it allows filtering 4CIF (704 × 576 pixels @30 fps) video sequences in real-time at frequencies lower than 10.16 Mhz.

Area-delay Trade-offs of Texture Decompressors for a Graphics Processing Unit

Graphics Processing Units have become a booster for the microelectronics industry. However, due to intellectual property issues, there is a serious lack of information on implementation details of the hardware architecture that is behind GPUs. For instance, the way texture is handled and decompressed in a GPU to reduce bandwidth usage has never been dealt with in depth from a hardware point of view. This work addresses a comparative study on the hardware implementation of different texture decompression algorithms for both conventional (PCs and video game consoles) and mobile platforms. Circuit synthesis is performed targeting both a reconfigurable hardware platform and a 90nm standard cell library. Area-delay trade-offs have been extensively analyzed, which allows us to compare the complexity of decompressors and thus determine suitability of algorithms for systems with limited hardware resources.

Run-time Scalable Architecture for Deblocking Filtering in H.264/AVC-SVC Video Codecs

Systems relying on fixed hardware components with a static level of parallelism can suffer from an underuse of logical resources, since they have to be designed for the worst-case scenario. This problem is especially important in video applications due to the emergence of new flexible standards, like Scalable Video Coding (SVC), which offer several levels of scalability. In this paper, Dynamic and Partial Reconfiguration (DPR) of modern FPGAs is used to achieve run-time variable parallelism, by using scalable architectures where the size can be adapted at run-time. Based on this proposal, a scalable Deblocking Filter core (DF), compliant with the H.264/AVC and SVC standards has been designed. This scalable DF allows run-time addition or removal of computational units working in parallel. Scalability is offered together with a scalable parallelization strategy at the macroblock (MB) level, such that when the size of the architecture changes, MB filtering order is modified accordingly

Scalable software architecture for on-line multi-camera video processing

In this paper we present a scalable software architecture for on-line multi-camera video processing, that guarantees a good trade off between computational power, scalability and flexibility. The software system is modular and its main blocks are the Processing Units (PUs), and the Central Unit. The Central Unit works as a supervisor of the running PUs and each PU manages the acquisition phase and the processing phase. Furthermore, an approach to easily parallelize the desired processing application has been presented. In this paper, as case study, we apply the proposed software architecture to a multi-camera system in order to efficiently manage multiple 2D object detection modules in a real-time scenario. System performance has been evaluated under different load conditions such as number of cameras and image sizes. The results show that the software architecture scales well with the number of camera and can easily works with different image formats respecting the real time constraints. Moreover, the parallelization approach can be used in order to speed up the processing tasks with a low level of overhead

A computer vision approach for weeds identification through Support Vector Machines

This paper outlines an automatic computervision system for the identification of avena sterilis which is a special weed seed growing in cereal crops. The final goal is to reduce the quantity of herbicide to be sprayed as an important and necessary step for precision agriculture. So, only areas where the presence of weeds is important should be sprayed. The main problems for the identification of this kind of weed are its similar spectral signature with respect the crops and also its irregular distribution in the field. It has been designed a new strategy involving two processes: image segmentation and decision making. The image segmentation combines basic suitable image processing techniques in order to extract cells from the image as the low level units. Each cell is described by two area-based attributes measuring the relations among the crops and weeds. The decision making is based on the SupportVectorMachines and determines if a cell must be sprayed. The main findings of this paper are reflected in the combination of the segmentation and the SupportVectorMachines decision processes. Another important contribution of this approach is the minimum requirements of the system in terms of memory and computation power if compared with other previous works. The performance of the method is illustrated by comparative analysis against some existing strategies.

Rational invariants on the space of all structures of algebras on a two-dimensional vector space

Rational invariants on the space of all structures of algebras on a two-dimensional vector space

CumuloNimbo: A highly-scalable transaction processing platform as a service.

One of the main challenges facing next generation Cloud platform services is the need to simultaneously achieve ease of programming, consistency, and high scalability. Big Data applications have so far focused on batch processing. The next step for Big Data is to move to the online world. This shift will raise the requirements for transactional guarantees. CumuloNimbo is a new EC-funded project led by Universidad Politécnica de Madrid (UPM) that addresses these issues via a highly scalable multi-tier transactional platform as a service (PaaS) that bridges the gap between OLTP and Big Data applications.

StreamCloud: An elastic and scalable data streaming system

Many applications in several domains such as telecommunications, network security, large scale sensor networks, require online processing of continuous data lows. They produce very high loads that requires aggregating the processing capacity of many nodes. Current Stream Processing Engines do not scale with the input load due to single-node bottlenecks. Additionally, they are based on static con?gurations that lead to either under or over-provisioning. In this paper, we present StreamCloud, a scalable and elastic stream processing engine for processing large data stream volumes. StreamCloud uses a novel parallelization technique that splits queries into subqueries that are allocated to independent sets of nodes in a way that minimizes the distribution overhead. Its elastic protocols exhibit low intrusiveness, enabling effective adjustment of resources to the incoming load. Elasticity is combined with dynamic load balancing to minimize the computational resources used. The paper presents the system design, implementation and a thorough evaluation of the scalability and elasticity of the fully implemented system.