Grâce aux progrès récents en photogrammétrie, il est désormais possible de reconstruire automatiquement un modèle d'une scène 3D à partir de photographies ou d'une vidéo. La reconstruction est réalisée en plusieurs étapes. Tout d'abord, on détecte des traits saillants (features) dans chaque image, souvent des points mais plus généralement des régions. Puis on cherche à les mettre en correspondance entre images. On utilise ensuite les traits communs à deux images pour déterminer la pose (positions et orientations) relative des images. Puis les poses sont mises dans un même repère global et la position des traits saillants dans l'espace est reconstruite (structure from motion). Enfin, un modèle 3D dense de la scène peut être estimé. La détection de traits saillants, leur appariement, ainsi que l'estimation de la position des caméras, jouent des rôles primordiaux dans la chaîne de reconstruction 3D. Des imprécisions ou des erreurs dans ces étapes ont un impact majeur sur la précision et la robustesse de la reconstruction de la scène entière. Dans cette thèse, nous nous intéressons à l'amélioration des méthodes pour établir la correspondance entre régions caractéristiques et pour les sélectionner lors de l'estimation des poses de caméras, afin de rendre les résultats de reconstruction plus robustes et plus précis. Nous introduisons tout d'abord une contrainte photométrique pour une paire de correspondances (VLD) au sein d'une même image, qui est plus fiable que les contraintes purement géométriques. Puis, nous proposons une méthode semi-locale (K-VLD) pour la mise en correspondance, basée sur cette contrainte photométrique. Nous démontrons que notre méthode est très robuste pour des scènes rigides, mais aussi non-rigides ou répétitives, et qu'elle permet d'améliorer la robustesse et la précision de méthodes d'estimation de poses, notamment basées sur RANSAC. Puis, pour améliorer l'estimation de la position des caméras, nous analysons la précision des reconstructions et des estimations de pose en fonction du nombre et de la qualité des correspondances. Nous en dérivons une formule expérimentale caractérisant la relation ``qualité contre quantité''. Sur cette base, nous proposons une méthode pour sélectionner un sous-ensemble des correspondances de meilleure qualité de façon à obtenir une très haute précision en estimation de poses. Nous cherchons aussi raffiner la précision de localisation des points en correspondance. Pour cela, nous développons une extension de la méthode de mise en correspondance aux moindres carrés (LSM) en introduisant un échantillonnage irrégulier et une exploration des échelles d'images. Nous montrons que le raffinement et la sélection de correspondances agissent indépendamment pour améliorer la reconstruction. Combinées, les deux méthodes produisent des résultats encore meilleurs / With the recent progress in photogrammetry, it is now possible to automatically reconstruct a model of a 3D scene from pictures or videos. The model is reconstructed in several stages. First, salient features (often points, but more generally regions) are detected in each image. Second, features that are common in images pairs are matched. Third, matched features are used to estimate the relative pose (position and orientation) of images. The global poses are then computed as well as the 3D location of these features (structure from motion). Finally, a dense 3D model can be estimated. The detection of salient features, their matching, as well as the estimation of camera poses play a crucial role in the reconstruction process. Inaccuracies or errors in these stages have a major impact on the accuracy and robustness of reconstruction for the entire scene. In this thesis, we propose better methods for feature matching and feature selection, which improve the robustness and accuracy of existing methods for camera position estimation. We first introduce a photometric pairwise constraint for feature matches (VLD), which is more reliable than geometric constraints. Then we propose a semi-local matching approach (K-VLD) using this photometric match constraint. We show that our method is very robust, not only for rigid scenes but also for non-rigid and repetitive scenes, which can improve the robustness and accuracy of pose estimation methods, such as based on RANSAC. To improve the accuracy in camera position estimation, we study the accuracy of reconstruction and pose estimation in function of the number and quality of matches. We experimentally derive a “quantity vs. quality” relation. Using this relation, we propose a method to select a subset of good matches to produce highly accurate pose estimations. We also aim at refining match position. For this, we propose an improvement of least square matching (LSM) using an irregular sampling grid and image scale exploration. We show that match refinement and match selection independently improve the reconstruction results, and when combined together, the results are further improved

Made available in DSpace on 2014-06-11T19:23:31Z (GMT). No. of bitstreams: 0 Previous issue date: 2003-06Bitstream added on 2014-06-13T20:50:28Z : No. of bitstreams: 1 nogueirajunior_jb_me_prud.pdf: 2403875 bytes, checksum: e914bf9771603f7fb120e3e7f9657040 (MD5) / O controle de qualidade (CQ) de dados cartográficos é uma atividade de extrema importância dentro das Ciências Cartográficas. Envolve, em geral, a análise de 7 elementos, dentre eles a qualidade posicional do produto cartográfico. No Brasil, a análise da qualidade posicional é baseada no PEC (Padrão de Exatidão Cartográfica), contida no Decreto Lei 89.817 de 20/06/1984, o qual foi concebido quando se tinha disponível apenas tecnologia para gerar produtos analógicos. A realidade hoje é outra, já que a cartografia digital tem praticamente substituído a analógica. Os objetivos principais deste trabalho são os seguintes: investigar os problemas relacionados ao controle de qualidade tendo em vista a cartografia digital, incluindo o Decreto Lei 89.817; apresentar um estudo detalhado sobre amostragem, visando identificar o número de pontos necessários no controle de qualidade posicional baseado em fundamentos científicos. A metodologia proposta consiste em aplicar estudos sobre amostragem e testes estatísticos de análise de tendência e precisão a dados disponíveis e com isso determinar procedimentos e rotinas ideais para avaliar, não só a qualidade posicional, mas também da linhagem, fidelidade de atributos, completeza, consistência lógica, fidelidade à semântica e temporalidade. / The quality control (CQ) of cartographic data is an activity of great value in the field of Cartographic Sciences. It involves, generally, the analysis of 7 elements, among them the positional quality of the cartographic product. In Brazil, it is based in Brazilian Cartographic Accuracy Standard contained in the Decree Law 89.817 of 06/20/1984, conceived in the moment that only analogical products were available. Nowadays, the reality is different, because the digital cartography has been substituting the analogical one. The main objectives of this work are the following ones: (a) to investigate the problems related for digital cartography’s quality control, including the Decree 89.817; (b) to present a detailed study on sampling, concerning to identify the amount points of necessary in the quality control based on scientific foundations. The proposed methodology consists in applying studies on sampling, statistical tests tendency and precision analysis for available data and, thereafter, to determine procedures and suitable routines for evaluating not only the positional quality but also lineage, atttribute accuracy, completeness, logical consistency, semantic accuracy and temporal information.

Made available in DSpace on 2014-10-09T12:41:35Z (GMT). No. of bitstreams: 0 / Made available in DSpace on 2014-10-09T13:56:14Z (GMT). No. of bitstreams: 0 / Dissertação (Mestrado) / IPEN/D / Instituto de Pesquisas Energeticas e Nucleares - IPEN-CNEN/SP

abstract: The purpose of this study was to examine the longitudinal relations of maternal behaviors, children`s temperamental negative emotionality, and children`s emotion perception processes, including emotion perception accuracy (EPA) and emotion perception bias (EPB), to children`s conduct disorder symptoms in a normative sample. Separate structural equation models were conducted to assess whether parenting or children`s proneness to negative emotions at 24-30 (T2), 36-42 (T3) and 48-54 (T4) months predicted children`s EPA and EPB over time, and whether T3 and T4 children`s emotion perception processes were predictive of children`s conduct disorder at 72 months of age (T5). None of the hypothesized longitudinal relations was supported; however, other noteworthy results were observed. T3 children`s proneness to negative emotions was positively related to children`s concurrent bias toward anger. The latent constructs of negative parenting, children`s proneness to negative emotions, and the observed measure of children`s emotion perception accuracy showed stability over time, whereas the observed measures of children`s bias toward understanding distinct negative emotions were unrelated across time. In addition, children`s expressive language was predicted by children`s earlier emotion perception accuracy, which emphasized the importance of improving children`s emotion understanding skills during early years. Furthermore, the previously established negative relation between EPA and EPB variables was only partially supported. Findings regarding the relations between parenting, children`s negative emotionality and emotion perception processes are discussed from a developmental perspective. / Dissertation/Thesis / M.S. Family and Human Development 2011

Fast and accurate monitoring of tropical forest disturbance is essential for understanding current patterns of deforestation as well as helping eliminate illegal logging. This dissertation explores the use of data from different satellites for near real-time monitoring of forest disturbance in tropical forests, including: development of new monitoring methods; development of new assessment methods; and assessment of the performance and operational readiness of existing methods. Current methods for accuracy assessment of remote sensing products do not address the priority of near real-time monitoring of detecting disturbance events as early as possible. I introduce a new assessment framework for near real-time products that focuses on the timing and the minimum detectable size of disturbance events. The new framework reveals the relationship between change detection accuracy and the time needed to identify events. In regions that are frequently cloudy, near real-time monitoring using data from a single sensor is difficult. This study extends the work by Xin et al. (2013) and develops a new time series method (Fusion2) based on fusion of Landsat and MODIS (Moderate Resolution Imaging Spectroradiometer) data. Results of three test sites in the Amazon Basin show that Fusion2 can detect 44.4% of the forest disturbance within 13 clear observations (82 days) after the initial disturbance. The smallest event detected by Fusion2 is 6.5 ha. Also, Fusion2 detects disturbance faster and has less commission error than more conventional methods. In a comparison of coarse resolution sensors, MODIS Terra and Aqua combined provides faster and more accurate detection of disturbance events than VIIRS (Visible Infrared Imaging Radiometer Suite) and MODIS single sensor data. The performance of near real-time monitoring using VIIRS is slightly worse than MODIS Terra but significantly better than MODIS Aqua. New monitoring methods developed in this dissertation provide forest protection organizations the capacity to monitor illegal logging events promptly. In the future, combining two Landsat and two Sentinel-2 satellites will provide global coverage at 30 m resolution every 4 days, and routine monitoring may be possible at high resolution. The methods and assessment framework developed in this dissertation are adaptable to newly available datasets.

In today’s world the internet and social media are omnipresent and information is accessible to everyone. This shifted the advantage from those who have access to information to those who do so first. Identifying new events as they emerge is of substantial value to financial institutions who consider realtime information in their decision making processes, as well as for journalists that report about breaking news and governmental agencies that collect information and respond to emergencies. First Story Detection is the task of identifying those documents in a stream of documents that talk about new events first. This seemingly simple task is non-trivial as the computational effort increases with every processed document. Standard approaches to solve First Story Detection determine a document’s novelty by comparing it to previously seen documents. This results in the highest reported accuracy but even the currently fastest system only scales to 10% of the Twitter stream. In this thesis, we propose a new algorithm family, called memory-based methods, able to scale to the full Twitter stream on a single core. Our memory-based method computes a document’s novelty up to two orders of magnitude faster than state-of-the-art systems without sacrificing accuracy. This thesis additional provides original work on the impact of processing unbounded data streams on detection accuracy. Our experiments reveal for the first time that the novelty scores of state-of-the-art comparison based and memory-based methods decay over time. We show how to counteract the discovered novelty decay and increase detection accuracy. Additionally, we show that memory-based methods are applicable beyond First Story Detection by building the first real time rumour detection system on social media streams.

Individual tree detection and modelling forest parameters using Airborne Laser Scanner data (Light Detection and Ranging (LiDAR) is becoming increasingly important for the monitoring and sustainable management of forests. Remote sensing has been a useful tool for individual tree analysis in the past decade, although inadequate spatial resolution from satellites means that only airborne systems have sufficient spatial resolution to conduct individual tree analysis. Moreover, recent advances in airborne LiDAR now provide high horizontal resolution as well as information in the vertical dimension. However, it is challenging to fully exploit and utilize small-footprint LiDAR data for detailed tree analysis. Procedures for forest biomass quantification and forest attributes measurement using LiDAR data have improved at a rapid pace as more robust and sophisticated modelling used to improve the studies. This thesis contains an evaluation of three approaches of utilizing LiDAR data for individual tree forest measurement. The first explores the relationship between LiDAR metrics and field reference to assess the correlation between LiDAR and field data at the individual-tree level. The intention was not to detect trees automatically, but to develop a LiDAR-AGB model based on trees that were mapped in the field so as to evaluate the relationships between LiDAR-type metrics under controlled conditions for the study sites, and field-derived AGB. A non-linear AGB model based on field data and LiDAR data was developed and LiDAR height percentile h80 and crown width measurement (CW) was found to best fit the data as evidenced by and Adj-R2 value of 0.63, the root mean squared error of the model of 14.8% and analysis of the residuals. This paper provides the foundation for a predictive LiDAR-AGB model at tree level over two study sites, Pasoh Forest Reserve and FRIM Forest Reserve. The second part of the thesis then takes this AGB-LiDAR relationship and combines it with individual tree crown delineation. This chapter shows the contribution of performing an automatic individual tree crown delineation over the wider forest areas. The individual tree crown delineation is composed of a five-step framework, which is unique in its automated determination of dominant crown sizes in a forest area and its adaption of the LiDAR-AGB model developed for the purpose of validation the method. This framework correctly delineated 84% and 88% of the tree crowns in the two forest study areas which is mostly dominated with lowland dipterocarp trees. Thirdly, parametric and non-parametric modelling approaches are proposed for modelling forest structural attributes. Selected modelling methods are compared for predicting 4 forest attributes, volume (V), basal area (BA), height (Ht) and aboveground biomass (AGB) at the species level. The AGB modelling in this paper is extracted using the LiDAR derived variables from the automated individual tree crown delineation, in contrast to the earlier AGB modelling where it is derived based on the trees that were mapped in the field. The selected non-parametric method included, k-nearest neighbour (k-NN) imputation methods: Most Similar Neighbour (MSN) and Gradient Nearest Neighbour (GNN), Random Forest (RF) and parametric approach: Ordinary Least Square (OLS) regression. To compare and evaluate these approaches a scaled root mean squared error (RMSE) between observed and predicted forest attribute sampled from both forest site was computed. The best method varied according to response variable and performance measure. OLS regression was to found to be the best performance method overall evidenced by RMSE after cross validation for BA (1.40 m2), V (1.03 m3), Ht (2.22 m) and AGB (96 Kg/tree) respectively, showed its applicability to wider conditions, while RF produced best overall results among the non-parametric methods tested. This thesis concludes with a discussion of the potential of LiDAR data as an independent source of important forest inventory data source when combined with appropriate designed sample plots in the field, and with appropriate modelling tools.

Este estudio tiene por objeto determinar en una primera parte, la capacidad predictiva des las redes neuronales tanto Rolling como Recursivas en la preedición de signos de la variación del precio del oro, para en una segunda parte determinar si los resultados obtenidos por una de estas redes es robusto en los distintos escenarios económicos o sea si se obtiene la misma predicción de signos bajo diferentes escenarios ficticios, los cuales se simulan con la técnica Bootstrap, para obtener una distribución de los retornos de la técnica.

This project has set out to design and optimise a robust and efficient interface for capillary electrophoresis-inductively coupled plasma mass spectrometry (CE-ICP-MS) and to investigate the application of the technique in elemental speciation studies. An interface was constructed using a commercial microconcentric nebuliser (MCN) and a cyclonic spray chamber. The cyclonic spray chamber was designed specifically to provide rapid sample response and washout and to minimise sample dispersion. Isoforms of the heavy metal binding protein, metallothionein, were separated and the bound metals detected to characterise the interface. Suction from the self-aspirating nebuliser was identified as the principal factor controlling electrophoretic resolution.

Projektet har genomförts i samarbete med Mora kommun som önskat att prova UAV som teknik för att bestämma markhöjder i ett bostads- och grönområde där mark kommer att schaktas vid en marksanering. Projektets syfte har varit att jämföra kvalitet på insamlade markhöjder från UAV med laserskanning och terrestra metoder samt att undersöka hur resultatet påverkas av olika flyghöjder. Med metoderna avvägning och inmätning med satellitteknik mättes sju flygstödspunkter och fyra mätpunkter in. Vid flygstödspunkterna placerades flygsignaler under flygningarna med UAV vilka skedde på 120 respektive 60 meters höjd över marken. Insamlade data bearbetades i Agisoft Photoscan 1.4.2 och med det framtagna punktmolnet kunde mätningar i modeller genomföras. Även insamling av höjddata från Lantmäteriets GSD Höjddata, grid 2+ data inhämtades. Därefter jämfördes koordinater för punkterna från bearbetningen med insamlade data. Resultatet visar på att standardavvikelsen i höjd vid 120 meters flyghöjd var 64 millimeter. Vid exkludering av mätpunkt 13, som var felaktigt inmätt, blev avvikelsen endast 29 millimeter. För 60 meters flyghöjd var standardavvikelsen 76 millimeter och med mätpunkt 13 exkluderad endast 24 millimeter. Dessa resultat förhåller sig till kraven för kvalitet enligt Lantmäteriets HMK-standardnivå 2, vilket är nivån för kartering av tätorter för bland annat detaljplaneläggning. / The objective of the study was to compare elevation data, from UAV photo, with NRTK, trigonometric leveling and airborne laser scanning. The concrete issues to investigate were: • What quality is achieved by data acquisition using UAV, compared to laser scanning and terrestrial measurement with NRTK? • How much does different flight altitudes affect the result? The aerial photos were taken at 60 and 120 meters above ground, using seven flight support points and four ground control points. Those were measured with NRTK and terrestrial measurement. Collected images have been processed, with Agisoft Photoscan 1.4.2, into models in which coordinates are retrieved for comparison. The coordinates were compared and showed a deviation of 24 to 76 millimeters which is an acceptable result for a HMK standard level 2.