1,720,955 research outputs found
Reconocimiento de acciones en video
No full textSe considera el problema de identificar automáticamente la acción ejecutada en un video. Dada una lista de posibles acciones (por ejemplo, correr, sentarse, aplaudir, etc.) y un video en el que se muestra un actor llevando a cabo una de ellas, se quiere contar con un algoritmo que permita reconocer la acción siendo ejecutada. Las áreas de aplicación para el reconocimiento de acciones son muy variadas, e incluyen la identificación de personas, la obtención automática de descripciones de videos, la vigilancia basada en el análisis de videos capturados por cámaras de seguridad, la interacción con computadoras mediante movimientos, etc. El problema es desafiante porque, por un lado, la apariencia de una misma acción puede variar mucho en distintos videos y, por otro, acciones distintas pueden tener una apariencia similar. En esta tesis se estudia un enfoque particular para el tema, que intenta reconocer la acción ejecutada en un video a partir de los descriptores de parches espacio-temporales extraídos del mismo en determinados puntos de interés. Se analizan algoritmos para cada uno de los 3 pasos principales del enfoque: la detección de puntos de interés, la obtención de descriptores y el reconocimiento de la acción a partir de los descriptores. Para todos los pasos, se implementaron algoritmos existentes. Para el tercer paso se desarrollaron, además, 2 nuevos algoritmos. Uno de ellos apunta, principalmente, a lograr una mejora en el tiempo consumido para determinar la acción presente en un video. En líneas generales, la estrategia seguida para conseguir dicha mejora consiste en organizar de una manera particular la base de datos de descriptores. El otro intenta mejorar la capacidad de clasificar correctamente un video mediante la utilización de varios tipos de descriptores por punto de interés. El enfoque elegido es evaluado utilizando 2 bases de datos de videos disponibles públicamente, obteniendo resultados muy alentadores.In this thesis, we will be considering the problem of automatically identifying an action performed in a video. Given a list of possible actions (e.g. running, sitting, clapping hands, etc.) and a video showing an actor performing any one of them, the goal is to produce an algorithm for the recognition of the action being performed. The fields of application for action recognition are varied, and include people recognition, automatically achieving video descriptions, vigilance based on the analysis of security cam-captured videos, interaction with computers via movement, etc. The problem is challenging because, on one hand, the appearance of an action can vary in different videos and, on the other hand, different actions can have similar appearance. In this thesis a specific approach to the subject will be taken, that of trying to recognize an action performed in a given video using descriptors of spatio-temporal patches extracted from certain interest points in that same video. Algorithms for each of the three main steps will be analyzed: detecting interest points, obtaining descriptors, and recognizing the action using those descriptors. For every step, existing algorithms were applied. For the third step two new algorithms were developed. The first algorithm aims at lowering the amount of time consumed in determining the action present in the video. Broadly, the strategy followed to achieve such improvement is that of organizing the descriptors data base in a particular fashion. The second algorithm intends to improve on the ability to correctly classify a video through the use of several types of descriptors for each interest point. The chosen approach to the subject is then evaluated by the use of two freely available video data bases, with very promising results.Fil: Ubalde, Sebastián. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales; Argentina
Transfer Learning Decision Forests for Gesture Recognition
Full text linkDecision forests are an increasingly popular tool in computer vision problems. Their advantages include high computational efficiency, state-of-the-art accuracy and multi-class support. In this paper, we present a novel method for transfer learning which uses decision forests, and we apply it to recognize gestures and characters. We introduce two mechanisms into the decision forest framework in order to transfer knowledge from the source tasks to a given target task. The first one is mixed information gain, which is a data-based regularizer. The second one is label propagation, which infers the manifold structure of the feature space. We show that both of them are important to achieve higher accuracy. Our experiments demonstrate improvements over traditional decision forests in the ChaLearn Gesture Challenge and MNIST data set. They also compare favorably against other state-of-the-art classifiers.Fil: Goussies, Norberto Adrián. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Computación; Argentina. Consejo Nacional de Investigaciones Científicas y Técnicas; ArgentinaFil: Ubalde, Sebastián. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Computación; Argentina. Consejo Nacional de Investigaciones Científicas y Técnicas; ArgentinaFil: Mejail, Marta Estela. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Computación; Argentina. Consejo Nacional de Investigaciones Científicas y Técnicas; Argentin
Going Beyond Counting First Authors in Author Co-citation Analysis
Full text linkThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Variations on the Author
Full text link“Variations on the Author” discusses two of Eduardo Coutinho’s recent films (Um Dia na Vida, from 2010, and Últimas Conversas, posthumously released in 2015) and their contribution to the general question of documentary authorship. The director’s filmography is characterized by a consistent yet self-effacing form of authorial self-inscription: Coutinho often features as an interviewer that rather than express opinions propels discourses; an interviewer that is good at listening. This mode of self-inscription characterizes him as an author who is not expressive but who is nonetheless markedly present on the screen. In Um Dia na Vida, however, Coutinho is completely absent form the image, while Últimas Conversas, on the contrary, includes a confessional prologue that moves the director from the margins to the center of his films. This article examines the ways in which these works stand out in the filmography of a director who offers new insights into the notion of cinematic authorship
Appropriate Similarity Measures for Author Cocitation Analysis
Full text linkWe provide a number of new insights into the methodological discussion about author cocitation analysis. We first argue that the use of the Pearson correlation for measuring the similarity between authors’ cocitation profiles is not very satisfactory. We then discuss what kind of similarity measures may be used as an alternative to the Pearson correlation. We consider three similarity measures in particular. One is the well-known cosine. The other two similarity measures have not been used before in the bibliometric literature. Finally, we show by means of an example that our findings have a high practical relevance.information science;Pearson correlation;cosine;similarity measure;author cocitation analysis
Action recognition in depth videos
Full text linkEl problema de reconocer automáticamente una acción llevada a cabo en un video está recibiendo mucha atención en la comunidad de visión por computadora, con aplicaciones que van desde el reconocimiento de personas hasta la interacción persona-computador. Podemos pensar al cuerpo humano como un sistema de segmentos rígidos conectados por articulaciones, y al movimiento del cuerpo como una transformación continua de la configuración espacial de dichos segmentos. La llegada de cámaras de profundidad de bajo costo hizo posible el desarrollo de un algoritmo de seguimiento de personas preciso y eficiente, que obtiene la ubicación 3D de varias articulaciones del esqueleto humano en tiempo real. Esta tesis presenta contribuciones al modelado de la evolución temporal de los esqueletos. El modelado de la evolución temporal de descriptores de esqueleto plantea varios desafíos. En primer lugar, la posición 3D estimada para las articulaciones suele ser imprecisa. En segundo lugar, las acciones humanas presentan gran variabilidad intra-clase. Esta variabilidad puede encontrarse no sólo en la configuración de los esqueletos por separado (por ejemplo, la misma acción da lugar a diferentes configuraciones para diestros y para zurdos) sino tambión en la dinámica de la acción: diferentes personas pueden ejecutar una misma acción a distintas velocidades; las acciones que involucran movimientos periódicos (como aplaudir) pueden presentar diferentes cantidades de repeticiones de esos movimientos; dos videos de la misma acción puede estar no-alineados temporalmente; etc. Por último, acciones diferentes pueden involucrar configuraciones de esqueleto y movimientos similares, dando lugar a un escenario de gran similaridad inter-clase. En este trabajo exploramos dos enfoques para hacer frente a estas dificultades. En el primer enfoque presentamos una extensión a Edit Distance on Real sequence (EDR), una medida de similaridad entre series temporales robusta y precisa. Proponemos dos mejoras clave a EDR: una función de costo suave para el alineamiento de puntos y un algoritmo de alineamiento modificado basado en el concepto de Instancia-a-Clase (I2C, por el término en inglés: Instance-to-Class). La función de distancia resultante tiene en cuenta el ordenamiento temporal de las secuencias comparadas, no requiere aprendizaje de parámetros y es altamente tolerante al ruido y al desfasaje temporal. Además, mejora los resultados de métodos no-paramótricos de clasificación de secuencias, sobre todo en casos de alta variabilidad intra-clase y pocos datos de entrenamiento. En el segundo enfoque, reconocemos que la cantidad de esqueletos discriminativos en una secuencia puede ser baja. Los esqueletos restantes pueden ser ruidosos, tener configuraciones comunes a varias acciones (por ejemplo, la configuración correspondiente a un esqueleto sentado e inmóvil) u ocurrir en instantes de tiempo poco comunes para la acción del video. Por lo tanto, el problema puede ser naturalmente encarado como uno de Aprendizaje Multi Instancia (MIL por el término en inglés Multiple Instance Learning). En MIL, las instancias de entrenamiento se organizan en conjuntos o bags. Cada bag de entrenamiento tiene asignada una etiqueta que indica la clase a la que pertenece. Un bag etiquetado con una determinada clase contiene instancias que son características de la clase, pero puede (y generalmente así ocurre) también contener instancias que no lo son. Siguiendo esta idea, representamos los videos como bags de descriptores de esqueleto con marcas de tiempo, y proponemos un framework basado en MIL para el reconocimiento de acciones. Nuestro enfoque resulta muy tolerante al ruido, la variabilidad intra-clase y la similaridad inter-clase. El framework propuesto es simple y provee un mecanismo claro para regular la tolerancia al ruido, a la poca alineación temporal y a la variación en las velocidades de ejecución. Evaluamos los enfoques presentados en cuatro bases de datos públicas capturadas con cámaras de profundidad. En todos los casos, se trata de bases desafiantes. Los resultados muestran una comparación favorable de nuestras propuestas respecto al estado del arte.The problem of automatically identifying an action performed in a video is receiving a great deal of attention in the computer vision community, with applications ranging from people recognition to human computer interaction. We can think the human body as an articulated system of rigid segments connected by joints, and human motion as a continuous transformation of the spatial arrangement of those segments. The arrival of low-cost depth cameras has made possible the development of an accurate and efficient human body tracking algorithm, that computes the 3D location of several skeleton joints in real time. This thesis presents contributions concerning the modeling of the skeletons temporal evolution. Modeling the temporal evolution of skeleton descriptors is a challenging task. First, the estimated location of the 3D joints are usually inaccurate. Second, human actions have large intra-class variability. This variability may be found not only in the spatial configuration of individual skeletons (for example, the same action involves different configurations for righthanded and left-handed people) but also on the action dynamics: different people have different execution speeds; actions with periodic movements (like clapping) may involve different numbers of repetitions; two videos of the same action may be temporally misaligned; etc. Finally, different actions may involve similar skeletal configurations, as well as similar movements, effectively yielding large inter-class similarity. We explore two approaches to the problem that aim at tackling this difficulties. In the first approach, we present an extension to the Edit Distance on Real sequence (EDR), a robust and accurate similarity measure between time series. We introduce two key improvements to EDR: a weighted matching scheme for the points in the series and a modified aligning algorithm based on the concept of Instance-to-Class distance. The resulting distance function takes into account temporal ordering, requires no learning of parameters and is highly tolerant to noise and temporal misalignment. Furthermore, it improves the results of non-parametric sequence classification methods, specially in cases of large intra-class variability and small training sets. In the second approach, we explicitly acknowledge that the number of discriminative skeletons in a sequence might be low. The rest of the skeletons might be noisy or too person-specific, have a configuration common to several actions (for example, a sit still configuration), or occur at uncommon frames. Thus, the problem can be naturally treated as a Multiple Instance Learning (MIL) problem. In MIL, training instances are organized into bags. A bag from a given class contains some instances that are characteristic of that class, but might (and most probably will) contain instances that are not. Following this idea, we represent videos as bags of time-stamped skeleton descriptors, and we propose a new MIL framework for action recognition from skeleton sequences. We found that our approach is highly tolerant to noise, intra-class variability and inter-class similarity. The proposed framework is simple and provides a clear way of regulating tolerance to noise, temporal misalignment and variations in execution speed. We evaluate the proposed approaches on four publicly available challenging datasets captured by depth cameras, and we show that they compare favorably against other state-of-the-art methods.Fil:Ubalde, Sebastián. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales; Argentina
Dispelling the Myths Behind First-author Citation Counts
Full text linkWe conducted a full-scale evaluative citation analysis study of scholars in the XML research field to explore just how different from each other author rankings resulting from different citation counting methods actually are, and to demonstrate the capability of emerging data and tools on the Web in supporting more realistic citation counting methods. Our results contest some common arguments for the continued
use of first-author citation counts in the evaluation of scholars, such as high correlations between author rankings by first-author citation counts and other citation
counting methods, and high costs of using more realistic citation counting methods that are not well-supported by the ISI databases. It is argued that increasingly available digital full text research papers make it possible for citation analysis studies to go beyond what the ISI databases have directly supported and to employ more
sophisticated methods
koamabayili/VECTRON-author-checklist: VECTRON author checklist
No full textWe have done our best to complete the author checklist relating to the use of animals in the hut study. Note that the objective for the hut study was to evaluate the IRS treatment applications for residual efficacy against Anopheles mosquitoes, including the local An. coluzzii mosquito population. Cows were only used to attract mosquitoes into the huts and no tests were carried out directly on the cows. The author checklist is intended for use with studies where experiments are carried out on animals, which is why we have had such difficulty in completing this for the hut study, as many of the questions do not relate to how the cows were used
- …
