Orientador: Aluir Porfírio Dal Poz / Resumo: Neste trabalho foi proposta uma metodologia para a extração de contornos de telhados de edifícios no espaço-objeto, a partir da integração de um estéreo par de imagens aéreas de alta resolução e modelos 3D aproximados de telhado obtidos a partir de dados de varredura a LASER. Um modelo matemático considerando as propriedades radiométricas e geométricas dos telhados foi formulado a fim de representar o contorno do telhado no espaçoimagem, tendo como base o modelo de contorno ativo Snake. Esse modelo foi então adaptado para descrever os contornos no espaço-objeto considerando um estéreo par de imagens aéreas. Finalmente, o polígono ótimo que representa um dado contorno do telhado foi determinado a partir da otimização, via Programação Dinâmica, da função de energia criada. A solução obtida é uma representação mais acurada para o correspondente contorno do modelo 3D do telhado. O método desenvolvido apresenta também mecanismos para realizar a compensação automática de três tipos de problemas comuns em ambientes urbanos e que podem prejudicar a extração automática de telhados: obstruções perspectivas causadas por edifícios elevados, obstruções diretas causadas por vegetação que se eleva acima do telhado e sombras adjacentes aos telhados, as quais podem ser confundidas com as bordas do telhado. Os experimentos foram realizados utilizando imagens aéreas com GSD ≈ 0,10 m e nuvem de pontos LASER com densidade média de 6 pontos/m2. Os resultados mostraram que o método funciona adequad... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: In this work a methodology was proposed for extracting building roof contours in the object-space, by integration of a high-resolution aerial images stereo pair and 3D roof models reconstructed from LASER scanning data. A mathematical model considering the radiometric and geometric properties of roofs was developed in order to represent the roof contour in the image-space, based on the Snake active contour model. Then, the model was adapted to represent the contours in the object space considering a stereo pair of aerial images. Finally, the optimal polygon representing a selected roof contour was obtained by optimizing the proposed energy function using Dynamic Programming algorithm. The solution obtained, i.e., a polygon representing each 3D roof contour, will be a higher accurate representation for the correspondent contour of the 3D roof model. The proposed method also presents mechanisms to perform the compensation of three types of common problems in urban environment and which can disturb the automatic roof extraction: perspective occlusions caused by high buildings, occlusions caused by vegetation that covers the roof and shadows that are adjacent to the roofs which can be misinterpreted as roof edges. The experiments were performed using aerial images with GSD ≈ 0,10 m and LASER point cloud with average density of 6 points/m2. The results showed that the proposed method works properly in contour extraction of roofs with occlusion and shadows areas, presenting complet... (Complete abstract click electronic access below) / Doutor

Indiana University-Purdue University Indianapolis (IUPUI) / Image is a dominant medium among video, 3D model, and other media for visualizing environment and creating virtual access on the Internet. The location of image capture is, however, subjective and has relied on the esthetic sense of photographers up until this point. In this paper, we will not only visualize areas with images, but also propose a general framework to determine where the most distinct viewpoints should be located. Starting from elevation data, we present spatial and content information in ground-based images such that (1) a given number of images can have maximum coverage on informative scenes; (2) a set of key views can be selected with certain continuity for representing the most distinct views. According to the scene visibility, continuity, and data redundancy, we evaluate viewpoints numerically with an object-emitting illumination model. Our key view exploration may eventually reduce the visual data to transmit, facilitate image acquisition, indexing and interaction, and enhance perception of spaces. Real sample images are captured based on planned positions to form a visual network to index the area.

<p>Forest canopies are a critical component of forest ecosystems as they influence many important functions. Specifically, the structure of forest canopies is a driver of the magnitude and rate of these functions. Therefore, being able to accurately measure canopy structure is crucial to ensure ecological models and forest management plans are as robust and efficient as possible. However, canopies are complex and dynamic entities and thus their structure can be challenging to accurately measure. Here we study the feasibility of using lidar to measure forest canopy structure across large spatial extents by investigating the compatibility of aerial and terrestrial lidar systems. Building on known structure-function relationships measured with terrestrial lidar, we establish grounds for scaling these relationships to the aerial scale. This would enable accurate measures of canopy structural complexity to be acquired at landscape and regional scales without the time and labor requirements of terrestrial data collection. Our results illustrate the potential for measures of canopy height, vegetation area, horizontal cover, and canopy roughness to be upscaled. Furthermore, we highlight the benefit of utilizing multivariate measures of canopy structure, and the capacity of lidar to identify forest structural types. Moving forward, lidar is a tool to be utilized in tandem with other technologies to best understand the spatial and temporal dynamics of forests and the influence of physical ecosystem structure. </p>

This dissertation addresses the difficulties of semantic segmentation when dealing with an extensive collection of images and 3D point clouds. Due to the ubiquity of digital cameras that help capture the world around us, as well as the advanced scanning techniques that are able to record 3D replicas of real cities, the sheer amount of visual data available presents many opportunities for both academic research and industrial applications. But the mere quantity of data also poses a tremendous challenge. In particular, the problem of distilling useful information from such a large repository of visual data has attracted ongoing interests in the fields of computer vision and data mining. Structural Semantics are fundamental to understanding both natural and man-made objects. Buildings, for example, are like languages in that they are made up of repeated structures or patterns that can be captured in images. In order to find these recurring patterns in images, I present an unsupervised frequent visual pattern mining approach that goes beyond co-location to identify spatially coherent visual patterns, regardless of their shape, size, locations and orientation. First, my approach categorizes visual items from scale-invariant image primitives with similar appearance using a suite of polynomial-time algorithms that have been designed to identify consistent structural associations among visual items, representing frequent visual patterns. After detecting repetitive image patterns, I use unsupervised and automatic segmentation of the identified patterns to generate more semantically meaningful representations. The underlying assumption is that pixels capturing the same portion of image patterns are visually consistent, while pixels that come from different backdrops are usually inconsistent. I further extend this approach to perform automatic segmentation of foreground objects from an Internet photo collection of landmark locations. New scanning technologies have successfully advanced the digital acquisition of large-scale urban landscapes. In addressing semantic segmentation and reconstruction of this data using LiDAR point clouds and geo-registered images of large-scale residential areas, I develop a complete system that simultaneously uses classification and segmentation methods to first identify different object categories and then apply category-specific reconstruction techniques to create visually pleasing and complete scene models.

La modélisation géométrique et la sémantisation de scènes intérieures à partir d'échantillon de points et un sujet de recherche qui prend de plus en plus d'importance. Cependant, le traitement d'un ensemble volumineux de données est rendu difficile d'une part par le nombre élevé d'objets parasitant la scène et d'autre part par divers défauts d'acquisitions comme par exemple des données manquantes ou un échantillonnage de la scène non isotrope. Cette thèse s'intéresse de près à de nouvelles méthodes permettant de modéliser géométriquement un nuage de point non structuré et d’y donner de la sémantique. Dans le chapitre 2, nous présentons deux méthodes permettant de transformer le nuage de points en un ensemble de formes. Nous proposons en premier lieu une méthode d'extraction de lignes qui détecte des segments à partir d'une coupe horizontale du nuage de point initiale. Puis nous introduisons une méthode par croissance de régions qui détecte et renforce progressivement des régularités parmi les formes planaires. Dans la première partie du chapitre 3, nous proposons une méthode basée sur de l'analyse statistique afin de séparer de la structure de la scène les objets la parasitant. Dans la seconde partie, nous présentons une méthode d'apprentissage supervisé permettant de classifier des objets en fonction d'un ensemble de formes planaires. Nous introduisons dans le chapitre 4 une méthode permettant de modéliser géométriquement le volume d'une pièce (sans meubles). Une formulation énergétique est utilisée afin de labelliser les régions d’une partition générée à partir de formes élémentaires comme étant intérieur ou extérieur de manière robuste au bruit et aux données. / Geometric modeling and semantization of indoor scenes from sampled point data is an emerging research topic. Recent advances in acquisition technologies provide highly accurate laser scanners and low-cost handheld RGB-D cameras for real-time acquisition. However, the processing of large data sets is hampered by high amounts of clutter and various defects such as missing data, outliers and anisotropic sampling. This thesis investigates three novel methods for efficient geometric modeling and semantization from unstructured point data: Shape detection, classification and geometric modeling. Chapter 2 introduces two methods for abstracting the input point data with primitive shapes. First, we propose a line extraction method to detect wall segments from a horizontal cross-section of the input point cloud. Second, we introduce a region growing method that progressively detects and reinforces regularities of planar shapes. This method utilizes regularities common to man-made architecture, i.e. coplanarity, parallelism and orthogonality, to reduce complexity and improve data fitting in defect-laden data. Chapter 3 introduces a method based on statistical analysis for separating clutter from structure. We also contribute a supervised machine learning method for object classification based on sets of planar shapes. Chapter 4 introduces a method for 3D geometric modeling of indoor scenes. We first partition the space using primitive shapes detected from permanent structures. An energy formulation is then used to solve an inside/outside labeling of a space partitioning, the latter providing robustness to missing data and outliers.

Submitted by Lucas Jamiro Barbosa (eng.lucasjb@gmail.com) on 2018-02-21T18:50:52Z No. of bitstreams: 1 barbosa_lj_me_prud.pdf: 5180173 bytes, checksum: 561a235397d5860ed83f8776cc40f6a4 (MD5) / Approved for entry into archive by ALESSANDRA KUBA OSHIRO ASSUNÇÃO (alessandra@fct.unesp.br) on 2018-02-21T19:37:41Z (GMT) No. of bitstreams: 1 barbosa_lj_me_prud.pdf: 5180173 bytes, checksum: 561a235397d5860ed83f8776cc40f6a4 (MD5) / Made available in DSpace on 2018-02-21T19:37:41Z (GMT). No. of bitstreams: 1 barbosa_lj_me_prud.pdf: 5180173 bytes, checksum: 561a235397d5860ed83f8776cc40f6a4 (MD5) Previous issue date: 2017-04-28 / Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) / O Sensoriamento Remoto tem-se mostrado, nos últimos anos, uma excelente ferramenta na aquisição de informações da cobertura da Terra. Dentre os diversos sensores remotos, o sistema de aquisição de dados por varredura LASER (Light Amplification by Stimulated Emission of Radiation) se apresenta como uma ferramenta poderosa na coleta de informações tridimensionais. A tecnologia lidar (Light Detection And Ranging), quando a bordo de aeronaves, pode ser denominada como Airborne Laser Scanning (ALS), diferente dos sistemas aerofotogramétricos de imageamento tradicionais, permite coletar, simultaneamente, pontos com coordenadas 3D sobre copas de árvores, bem como o terreno abaixo dela, em função da capacidade de registro de múltiplos retornos e da divergência do pulso. Por esta razão, esta tecnologia tem sido utilizada em diversas aplicações florestais, como manejo e recuperação florestal, silvicultura, exploração madeireira, dentre outras. Diversas pesquisas têm mostrado a possibilidade de utilização dos dados lidar na extração e delimitação de indivíduos arbóreos e, portanto, na estimativa de variáveis dendrométricas. Desta forma, o desenvolvimento de técnicas que proporcionem a automatização no delineamento das copas das árvores e na estimativa destas informações é de grande interesse. Contudo, grande parte das pesquisas relacionadas à detecção de árvores, delimitação de copa e estimativa de algumas variáveis são desenvolvidas considerando cenários homogêneos e específicos, onde a vegetação é caracterizada pela presença de árvores coníferas e/ou deciduais, ou florestas de exploração madeireira. Portanto, o objetivo desta pesquisa foi a implementação e avaliação de uma técnica que permita detectar indivíduos arbóreos e aglomerados, considerando um cenário urbano heterogêneo e complexo; e, destes indivíduos, estimar variáveis dendrométricas como área da copa; altura da árvore; e raio médio da copa. A metodologia proposta é realizada em três etapas e se baseia no uso do método de crescimento de regiões, aplicado à nuvem de pontos originais, ordenados quanto à altura. Além disso, são usados polígonos convexos visando a extração de indivíduos arbóreos e aglomerados. Para isso, são utilizados três parâmetros: distância mínima, buffer e perímetro comum. Foram realizados experimentos considerando dados reais e cenários diferentes em uma área urbana, para diferentes conjuntos de parâmetros utilizados no processo de delimitação das copas. Os mesmos foram avaliados quanto a acurácia temática, completeza e F-Score, calculados em função de referências obtidas de forma manual. Na delimitação de indivíduos arbóreos e aglomerados, simultaneamente, o maior valor de F-Score foi de 54% e na delimitação de indivíduos e aglomerados, em separado, o melhor resultado obtido foi 74% e 39%, respectivamente. Embora melhorias possam ser feitas visando aumentar estes indicadores, principalmente para aglomerados, pode-se considerar que o método proposto tem potencial de aplicação, sobretudo quando se tem por objetivo a extração de árvores individuais em ambiente urbano. / Remote Sensing has shown to be, in the last years, an excellent way of acquiring Earth’s surface data. Among all remote sensors, the system of data acquisition by LASER (Light Amplification by Stimulated Emission of Radiation) scanning has been presented as a powerful tool for three-dimensional information collection. lidar (Light Detection And Ranging) technology, when onboard of airplanes can be named Airborne Laser Scanning (ALS) and, differently from traditional photogrammetric techniques, allows the collection of simultaneously 3D points over tree crowns, as well as the ground underneath it, due to the recording capacity of multiple echoes arising from the divergence of the pulse. For this reason, this technology has been used in many different forest applications, as management and forest recovery, forestry, logging and others. Some researchers have shown the possibility of using lidar data on tree extraction and crown delineation and, therefore, on the estimation of their dendrometric variables. In this way, the development of techniques that can provide the automation of tree crowns delineation and estimation of this information has increased. However, most of the researches performed ever related to tree detection, canopy delineation and estimation of some dendrometric variables are developed considering homogeneous and specific scenarios where the vegetation is characterized by the presence of coniferous and/or deciduous trees. For this purpose, the objective of this research was the implementation and evaluation of a technique capable of detecting individual trees and clumps, considering a heterogeneous urban scenario. Additionally, from those individual trees some dendrometric variables such as crown area; tree height and average crown radius were estimated. Experiments were conducted considering different study areas in an urban environment varying the parameters used in the crown delineation process. Those experiments were evaluated in terms of thematic accuracy, completeness and F-Score, computed based on reference values obtained manually. When the simultaneous delimitation of arboreal individuals and agglomerates was performed the best F-Score was 54%. For independent processing, the best result was 74% and 39%, respectively, for individuals and agglomerates. Although improvements can be performed aiming to improve those indicators, mainly to clumps, it is possible to consider that the proposed method has potential, especially when the objective is the extraction of individual trees in an urban environment.

3D semantic segmentation is an expanding topic within the field of computer vision, which has received more attention in recent years due to the development of more powerful GPUs and the newpossibilities offered by deep learning techniques. Simultaneously, the amount of available spatial LiDAR data over Sweden has also increased. This work combines these two advances and investigates if a 3D deep learning model for semantic segmentation can learn to detect terrain roughness in airborne LiDAR data. The annotations for terrain roughness used in this work are taken from SGUs 2D soil type map. Other airborne data sources are also used to filter the annotations and see if additional information can boost the performance of the model.  Since this is the first known attempt at terrain roughness classification from 3D data, an initial test was performed where fields were classified. This ensured that the model could process airborne LiDAR data and work for a terrain classification task. The classification of fields showed very promising results without any fine-tuning. The results for the terrain roughness classification task show that the model could find a pattern in the validation data but had difficulty generalizing it to the test data. The filtering methods tested gave an increased mIoU and indicated that better annotations might be necessary to distinguish terrain roughness from other terrain types. None of the features obtained from the other data sources improved the results and showed no discriminating abilities when examining their individual histograms. In the end, more research is needed to determine whether terrain roughness can be detected from LiDAR data or not.

Au cours des dernières années, la cartographie urbaine 3D a suscité un intérêt croissant pour répondre à la demande d’applications d’analyse des scènes urbaines tournées vers un large public. Conjointement les techniques d’acquisition de données 3D progressaient. Les travaux concernant la modélisation et la visualisation 3D des villes se sont donc intensifiés. Des applications fournissent au plus grand nombre des visualisations efficaces de modèles urbains à grande échelle sur la base des imageries aérienne et satellitaire. Naturellement, la demande s’est portée vers des représentations avec un point de vue terrestre pour offrir une visualisation 3D plus détaillée et plus réaliste. Intégrées dans plusieurs navigateurs géographiques comme Google Street View, Microsoft Visual Earth ou Géoportail, ces modélisations sont désormais accessibles et offrent une représentation réaliste du terrain, créée à partir des numérisateurs mobiles terrestres. Dans des environnements urbains, la qualité des données obtenues à partir de ces véhicules terrestres hybrides est largement entravée par la présence d’objets temporairement statiques ou dynamiques (piétons, voitures, etc.) dans la scène. La mise à jour de la cartographie urbaine via la détection des modifications et le traitement des données bruitées dans les environnements urbains complexes, l’appariement des nuages de points au cours de passages successifs, voire la gestion des grandes variations d’aspect de la scène dues aux conditions environnementales constituent d’autres problèmes délicats associés à cette thématique. Plus récemment, les tâches de perception s’efforcent également de mener une analyse sémantique de l’environnement urbain pour renforcer les applications intégrant des cartes urbaines 3D. Dans cette thèse, nous présentons un travail supportant le passage à l’échelle pour la cartographie 3D urbaine automatique incorporant la reconnaissance et l’intégration temporelle. Nous présentons en détail les pratiques actuelles du domaine ainsi que les différentes méthodes, les applications, les technologies récentes d’acquisition des données et de cartographie, ainsi que les différents problèmes et les défis qui leur sont associés. Le travail présenté se confronte à ces nombreux défis mais principalement à la classification des zones urbaines l’environnement, à la détection automatique des changements, à la mise à jour efficace de la carte et l’analyse sémantique de l’environnement urbain. Dans la méthode proposée, nous effectuons d’abord la classification de l’environnement urbain en éléments permanents et temporaires. Les objets classés comme temporaire sont ensuite retirés du nuage de points 3D laissant une zone perforée dans le nuage de points 3D. Ces zones perforées ainsi que d’autres imperfections sont ensuite analysées et progressivement éliminées par une mise à jour incrémentale exploitant le concept de multiples passages. Nous montrons que la méthode d’intégration temporelle proposée permet également d’améliorer l’analyse sémantique de l’environnement urbain, notamment les façades des bâtiments. Les résultats, évalués sur des données réelles en utilisant différentes métriques, démontrent non seulement que la cartographie 3D résultante est précise et bien mise à jour, qu’elle ne contient que les caractéristiques permanentes exactes et sans imperfections, mais aussi que la méthode est également adaptée pour opérer sur des scènes urbaines de grande taille. La méthode est adaptée pour des applications liées à la modélisation et la cartographie du paysage urbain nécessitant une mise à jour fréquente de la base de données. / Over the years, 3D urban cartography has gained widespread interest and importance in the scientific community due to an ever increasing demand for urban landscape analysis for different popular applications, coupled with advances in 3D data acquisition technology. As a result, in the last few years, work on the 3D modeling and visualization of cities has intensified. Lately, applications have been very successful in delivering effective visualizations of large scale models based on aerial and satellite imagery to a broad audience. This has created a demand for ground based models as the next logical step to offer 3D visualizations of cities. Integrated in several geographical navigators, like Google Street View, Microsoft Visual Earth or Geoportail, several such models are accessible to large public who enthusiastically view the real-like representation of the terrain, created by mobile terrestrial image acquisition techniques. However, in urban environments, the quality of data acquired by these hybrid terrestrial vehicles is widely hampered by the presence of temporary stationary and dynamic objects (pedestrians, cars, etc.) in the scene. Other associated problems include efficient update of the urban cartography, effective change detection in the urban environment and issues like processing noisy data in the cluttered urban environment, matching / registration of point clouds in successive passages, and wide variations in environmental conditions, etc. Another aspect that has attracted a lot of attention recently is the semantic analysis of the urban environment to enrich semantically 3D mapping of urban cities, necessary for various perception tasks and modern applications. In this thesis, we present a scalable framework for automatic 3D urban cartography which incorporates recognition and temporal integration. We present in details the current practices in the domain along with the different methods, applications, recent data acquisition and mapping technologies as well as the different problems and challenges associated with them. The work presented addresses many of these challenges mainly pertaining to classification of urban environment, automatic change detection, efficient updating of 3D urban cartography and semantic analysis of the urban environment. In the proposed method, we first classify the urban environment into permanent and temporary classes. The objects classified as temporary are then removed from the 3D point cloud leaving behind a perforated 3D point cloud of the urban environment. These perforations along with other imperfections are then analyzed and progressively removed by incremental updating exploiting the concept of multiple passages. We also show that the proposed method of temporal integration also helps in improved semantic analysis of the urban environment, specially building façades. The proposed methods ensure that the resulting 3D cartography contains only the exact, accurate and well updated permanent features of the urban environment. These methods are validated on real data obtained from different sources in different environments. The results not only demonstrate the efficiency, scalability and technical strength of the method but also that it is ideally suited for applications pertaining to urban landscape modeling and cartography requiring frequent database updating.

<p dir="ltr">As a new spaceborne laser remote sensing system, the Global Ecosystem Dynamics Investigation, or GEDI, is being widely used for monitoring forest ecosystems. However, its measurements are subject to uncertainties that will affect the calculation of ground elevation and vegetation height. This research intends to investigate the quality of the GEDI elevation data and its relevance to topography and land cover.</p><p dir="ltr">In this study, the elevation of the GEDI data is compared to 3DEP DEM, which has a higher resolution and accuracy. All the experiments in this study are conducted for two locations with vastly different terrain and land cover conditions, namely Tippecanoe County in Indiana and Mendocino County in California. Through this investigation we expect to gain a comprehensive understanding of GEDI’s elevation quality in various terrain and land cover conditions.</p><p dir="ltr">The results show that GEDI data in Tippecanoe County has better elevation accuracy than the GEDI data in Mendocino County. GEDI in Tippecanoe County is almost four times more accurate than in Mendocino County. Regarding land cover, GEDI have better accuracy in low vegetation areas than in forest areas. The ratio can be around three times better in Tippecanoe County and around one and half times better in Mendocino County. In terms of slope, GEDI data shows a clear positive correlation between RMSE and slope. The trend indicates as slope increases, the RMSE increases concurrently. In other words, slope and GEDI elevation accuracy are inversely related. In the experiment involving slope and land cover, the results show that slope is the most influential factor to GEDI elevation accuracy.</p><p dir="ltr">This study informs GEDI users of the factors they must consider for forest biomass calculation and topographic mapping applications. When high terrain slope and/or high vegetation is present, the GEDI data should be checked with other data sources like 3DEP DEM or any ground truth measurements to assure its quality. We expect these findings can help worldwide users understand that the quality of GEDI data is variable and dependent on terrain relief and land cover.</p>

L'analyse et la reconstruction automatique de scène urbaine 3D est un problème fondamental dans le domaine de la vision par ordinateur et du traitement numérique de la géométrie. Cette thèse présente des méthodologies pour résoudre le problème complexe de la reconstruction d'éléments urbains en 3D à partir de données aériennes Lidar ou bien de maillages générés par imagerie Multi-View Stereo (MVS). Nos approches génèrent une représentation précise et compacte sous la forme d'un maillage 3D comportant une sémantique de l'espace urbain. Deux étapes sont nécessaires ; une identification des différents éléments de la scène urbaine, et une modélisation des éléments sous la forme d'un maillage 3D. Le Chapitre 2 présente deux méthodes de classifications des éléments urbains en classes d'intérêts permettant d'obtenir une compréhension approfondie de la scène urbaine, et d'élaborer différentes stratégies de reconstruction suivant le type d'éléments urbains. Cette idée, consistant à insérer à la fois une information sémantique et géométrique dans les scènes urbaines, est présentée en détails et validée à travers des expériences. Le Chapitre 3 présente une approche pour détecter la 'Végétation' incluses dans des données Lidar reposant sur les processus ponctuels marqués, combinée avec une nouvelle méthode d'optimisation. Le Chapitre 4 décrit à la fois une approche de maillage 3D pour les 'Bâtiments' à partir de données Lidar et de données MVS. Des expériences sur des structures urbaines larges et complexes montrent les bonnes performances de nos systèmes. / Analysis and 3D reconstruction of urban scenes from physical measurements is a fundamental problem in computer vision and geometry processing. Within the last decades, an important demand arises for automatic methods generating urban scenes representations. This thesis investigates the design of pipelines for solving the complex problem of reconstructing 3D urban elements from either aerial Lidar data or Multi-View Stereo (MVS) meshes. Our approaches generate accurate and compact mesh representations enriched with urban-related semantic labeling.In urban scene reconstruction, two important steps are necessary: an identification of the different elements of the scenes, and a representation of these elements with 3D meshes. Chapter 2 presents two classification methods which yield to a segmentation of the scene into semantic classes of interests. The beneath is twofold. First, this brings awareness of the scene for better understanding. Second, deferent reconstruction strategies are adopted for each type of urban elements. Our idea of inserting both semantical and structural information within urban scenes is discussed and validated through experiments. In Chapter 3, a top-down approach to detect 'Vegetation' elements from Lidar data is proposed using Marked Point Processes and a novel optimization method. In Chapter 4, bottom-up approaches are presented reconstructing 'Building' elements from Lidar data and from MVS meshes. Experiments on complex urban structures illustrate the robustness and scalability of our systems.