Publications of Florent Poux
Bookmark and Share    
Full Text
See detailAn extension of CityJSON to support point clouds
Nys, Gilles-Antoine ULiege; Kharroubi, Abderrazzaq ULiege; Poux, Florent ULiege et al

in International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences (2021, June 30), XLII-B4-2021

The combination between dense point clouds and 3D vector objects permits new cartographic representation of urban information. This paper proposes an extension for the CityJSON encoding to support point ... [more ▼]

The combination between dense point clouds and 3D vector objects permits new cartographic representation of urban information. This paper proposes an extension for the CityJSON encoding to support point clouds. Following the 3.0 CityGML specifications, attributes and features are added to the core module of v1.0.1 CityJSON. Two solutions are proposed: inline complex geometries and external link to a remote file. The extended schema can be illustrated in four scenarios: detailed features visualization, fall-back solution in features reconstruction processes, simulating urban climate represented as vector fields, and true-to-life representation solution for complex elements such as solitary vegetation objects. It permits 3D city modelers to handle points clouds in a native way reducing files size and avoiding redundancy. All developments and documentation are available open-source. [less ▲]

Detailed reference viewed: 40 (2 ULiège)
Full Text
See detailTESSERAE3D: A benchmark for tesserae semantic segmentation in 3D point clouds
Kharroubi, Abderrazzaq ULiege; Van Wersch, Line ULiege; Billen, Roland ULiege et al

in ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences (2021, June 17), V-2-2021

3D point cloud of mosaic tesserae is used by heritage researchers, restorers, and archaeologists for digital investigations. Information extraction, pattern analysis, and semantic assignment are necessary ... [more ▼]

3D point cloud of mosaic tesserae is used by heritage researchers, restorers, and archaeologists for digital investigations. Information extraction, pattern analysis, and semantic assignment are necessary to complement geometric information. Automated processes that can speed up the task are highly sought after, especially new supervised approaches. However, the availability of labeled data necessary for training supervised learning models is a significant constraint. This paper introduces Tesserae3D, a 3D point cloud benchmark dataset for training and evaluating machine learning models, applied to mosaic tesserae segmentation. It is a publicly available, very high density and colored dataset, accompanied by a standard multi-class semantic segmentation baseline. It consists of about 502 million points and contains 11 semantic classes covering a wide range of tesserae types. We propose a semantic segmentation baseline building on radiometric and covariance features fed to ensemble learning methods. The results delineate an achievable 89% F1 score and are made available under https://github.com/akharroubi/Tesserae3D, providing a simple interface to improve the score based on feedback from the research community. [less ▲]

Detailed reference viewed: 32 (2 ULiège)
Full Text
See detailHow to automate LiDAR point cloud sub-sampling with Python
Poux, Florent ULiege

Article for general public (2020)

In this article, I will give you my two favourite 3D processes for quickly structuring and sub-sampling point cloud data with python. You will also be able to automate, export, visualize and integrate ... [more ▼]

In this article, I will give you my two favourite 3D processes for quickly structuring and sub-sampling point cloud data with python. You will also be able to automate, export, visualize and integrate results into your favourite 3D software, without any coding experience. I will focus on code optimization while using a minimum number of libraries (mainly NumPy) so that you can extend what you learnt with very high flexibility! [less ▲]

Detailed reference viewed: 110 (1 ULiège)
Full Text
See detailExtraction et Gestion Automatique de la sémantique dans les nuages de points
Poux, Florent ULiege

Scientific conference (2020, November 05)

Dans un contexte de capture de la réalité couplée à l’intelligence artificielle, la présentation s’intéresse au potentiel de recherche passionnant de traitements avancés sur nuages de points 3D, de leur ... [more ▼]

Dans un contexte de capture de la réalité couplée à l’intelligence artificielle, la présentation s’intéresse au potentiel de recherche passionnant de traitements avancés sur nuages de points 3D, de leur structuration, leur sémantique et les traitements non supervisés. Le travail présentera, entre autres, un modèle interopérable pour la gestion des nuages de points volumineux tout en intégrant la sémantique. [less ▲]

Detailed reference viewed: 41 (2 ULiège)
Full Text
See detailNuages de points, segments, sémantique et automatisation
Poux, Florent ULiege

Scientific conference (2020, October 20)

Challenges towards the recognition of debris in 3D environments

Detailed reference viewed: 33 (5 ULiège)
Full Text
See detailA Built Heritage Information System Based on Point Cloud Data: HIS-PC
Poux, Florent ULiege; Billen, Roland ULiege; Kasprzyk, Jean-Paul ULiege et al

in ISPRS International Journal of Geo-Information (2020), 9(10), 588

The digital management of an archaeological site requires to store, organise, access and represent all the information that is collected on the field. Heritage building information modelling ... [more ▼]

The digital management of an archaeological site requires to store, organise, access and represent all the information that is collected on the field. Heritage building information modelling, archaeological or heritage information systems now tend to propose a common framework where all the materials are managed from a central database and visualised through a 3D representation. In this research, we offer the development of a built heritage information system prototype based on a high-resolution 3D point cloud data set. The particularity of the approach is to consider a user-centred development methodology while avoiding meshing/down-sampling operations. The proposed system is initiated by a close collaboration between multi-modal users (managers, visitors, curators) and a development team (designers, developers, architects). The developed heritage information system permits the management of spatial and temporal information, including a wide range of semantics using relational along with NoSQL databases. The semantics used to describe the artifacts are subject to conceptual modelling. Finally, the system proposes a bi-directional communication with a 3D interface able to stream massive point clouds, which is a big step forward to provide a comprehensive site representation for stakeholders while minimising modelling costs. [less ▲]

Detailed reference viewed: 77 (9 ULiège)
Full Text
See detailUnsupervised segmentation of indoor 3D point cloud: application to object-based classification
Poux, Florent ULiege; Mattes, Christian; Kobbelt, Leif

in International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences (2020, September 03), XLIV-4(W1-2020), 111-118

Abstract. Point cloud data of indoor scenes is primarily composed of planar-dominant elements. Automatic shape segmentation is thus valuable to avoid labour intensive labelling. This paper provides a ... [more ▼]

Abstract. Point cloud data of indoor scenes is primarily composed of planar-dominant elements. Automatic shape segmentation is thus valuable to avoid labour intensive labelling. This paper provides a fully unsupervised region growing segmentation approach for efficient clustering of massive 3D point clouds. Our contribution targets a low-level grouping beneficial to object-based classification. We argue that the use of relevant segments for object-based classification has the potential to perform better in terms of recognition accuracy, computing time and lowers the manual labelling time needed. However, fully unsupervised approaches are rare due to a lack of proper generalisation of user-defined parameters. We propose a self-learning heuristic process to define optimal parameters, and we validate our method on a large and richly annotated dataset (S3DIS) yielding 88.1\% average F1-score for object-based classification. It permits to automatically segment indoor point clouds with no prior knowledge at commercially viable performance and is the foundation for efficient indoor 3D modelling in cluttered point clouds. [less ▲]

Detailed reference viewed: 373 (11 ULiège)
Full Text
See detailCityJSON Building Generation from Airborne LiDAR 3D Point Clouds
Nys, Gilles-Antoine ULiege; Poux, Florent ULiege; Billen, Roland ULiege

in ISPRS International Journal of Geo-Information (2020), 9(521),

The relevant insights provided by 3D City models greatly improve Smart Cities and their management policies. In the urban built environment, buildings frequently represent the most studied and modeled ... [more ▼]

The relevant insights provided by 3D City models greatly improve Smart Cities and their management policies. In the urban built environment, buildings frequently represent the most studied and modeled features. CityJSON format proposes a lightweight and developer-friendly alternative to CityGML. This paper proposes an improvement to the usability of 3D models providing an automatic generation method in CityJSON, to ensure compactness, expressivity, and interoperability. In addition to a compliance rate in excess of 92% for geometry and topology, the generated model allows the handling of contextual information, such as metadata and refined levels of details (LoD), in a built-in manner. By breaking down the building-generation process, it creates consistent building objects from the unique source of Light Detection and Ranging (LiDAR) point clouds. [less ▲]

Detailed reference viewed: 57 (10 ULiège)
Full Text
See detailAutomatic 3D Buildings Compact Reconstruction from LiDAR point clouds
Nys, Gilles-Antoine ULiege; Billen, Roland ULiege; Poux, Florent ULiege

in International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences (2020, August 12), (XLIII-B2-2020), 473-478

Point clouds generated from aerial LiDAR and photogrammetric techniques are great ways to obtain valuable spatial insights over large scale. However, their nature hinders the direct extraction and sharing ... [more ▼]

Point clouds generated from aerial LiDAR and photogrammetric techniques are great ways to obtain valuable spatial insights over large scale. However, their nature hinders the direct extraction and sharing of underlying information. The generation of consistent large-scale 3D city models from this real-world data is a major challenge. Specifically, the integration in workflows usable by decision-making scenarios demands that the data is structured, rich and exchangeable. CityGML permits new advances in terms of interoperable endeavour to use city models in a collaborative way. Efforts have led to render good-looking digital twins of cities but few of them take into account their potential use in finite elements simulations (wind, floods, heat radiation model, etc.). In this paper, we target the automatic reconstruction of consistent 3D city buildings highlighting closed solids, coherent surface junctions, perfect snapping of vertices, etc. It specifically investigates the topological and geometrical consistency of generated models from aerial LiDAR point cloud, formatted following the CityJSON specifications. These models are then usable to store relevant information and provides geometries usable within complex computations such as computational fluid dynamics, free of local inconsistencies (e.g. holes and unclosed solids). [less ▲]

Detailed reference viewed: 101 (9 ULiège)
Full Text
See detailMARKER-LESS MOBILE AUGMENTED REALITY APPLICATION FOR MASSIVE 3D POINT CLOUDS AND SEMANTICS
Kharroubi, Abderrazzaq ULiege; Billen, Roland ULiege; Poux, Florent ULiege

in International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences (2020, August 12), XLIII(B2), 255261

Mobile Augmented Reality (MAR) attracts significant research and development efforts from both the industry and academia, but rarely integrate massive 3D dataset’s interactions. The emergence of dedicated ... [more ▼]

Mobile Augmented Reality (MAR) attracts significant research and development efforts from both the industry and academia, but rarely integrate massive 3D dataset’s interactions. The emergence of dedicated AR devices and powerful Software Development Kit (ARCore for android and ARKit for iOS) improves performance on mobile devices (Smartphones and tablets). This is aided by new sensor integration and advances in computer vision that fuels the development of MAR. In this paper, we propose a direct integration of massive 3D point clouds with semantics in a web-based marker-less mobile Augmented Reality (AR) application for real-time visualization. We specifically investigate challenges linked to point cloud data structure and semantic injection. Our solution consolidates some of the overarching principles of AR, of which pose estimation, registration and 3D tracking. The developed AR system is tested on mobile phones web-browsers providing clear insights on the performance of the system. Promising results highlight a number of frame per second varying between 27 and 60 for a real-time point budget of 4.3 million points. The point cloud tested is composed of 29 million points and shows how our indexation strategy permits the integration of massive point clouds aiming at the point budget. The results also gives research directions concerning the dependence and delay related to the quality of the network connection, and the battery consumption since portable sensors are used all the time. [less ▲]

Detailed reference viewed: 145 (11 ULiège)
Full Text
See detailSELF-LEARNING ONTOLOGY FOR INSTANCE SEGMENTATION OF 3D INDOOR POINT CLOUD
Poux, Florent ULiege; Ponciano, Jean-Jacques

in International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences (2020, August 12), XLIII(B2), 309316

Automation in point cloud data processing is central for efficient knowledge discovery. In this paper, we propose an instance segmentation framework for indoor buildings datasets. The process is built on ... [more ▼]

Automation in point cloud data processing is central for efficient knowledge discovery. In this paper, we propose an instance segmentation framework for indoor buildings datasets. The process is built on an unsupervised segmentation followed by an ontology-based classification reinforced by self-learning. We use both shape-based features that only leverages the raw X, Y, Z attributes as well as relationship and topology between voxel entities to obtain a 3D structural connectivity feature describing the point cloud. These are then used through a planar-based unsupervised segmentation to create relevant clusters constituting the input of the ontology of classification. Guided by semantic descriptions, the object characteristics are modelled in an ontology through OWL2 and SPARQL to permit structural elements classification in an interoperable fashion. The process benefits from a self-learning procedure that improves the object description iteratively in a fully autonomous fashion. Finally, we benchmark the approach against several deep-learning methods on the S3DIS dataset. We highlight full automation, good performances, easy-integration and a precision of 99.99% for planar-dominant classes outperforming state-of-the-art deep learning. [less ▲]

Detailed reference viewed: 469 (7 ULiège)
Full Text
See detailInitial User-Centered Design of a Virtual Reality Heritage System: Applications for Digital Tourism
Poux, Florent ULiege; Valembois, Quentin ULiege; Mattes, Christian et al

in Remote Sensing (2020), 12(16), 2583

Reality capture allows for the reconstruction, with a high accuracy, of the physical reality of cultural heritage sites. Obtained 3D models are often used for various applications such as promotional ... [more ▼]

Reality capture allows for the reconstruction, with a high accuracy, of the physical reality of cultural heritage sites. Obtained 3D models are often used for various applications such as promotional content creation, virtual tours, and immersive experiences. In this paper, we study new ways to interact with these high-quality 3D reconstructions in a real-world scenario. We propose a user-centric product design to create a virtual reality (VR) application specifically intended for multi-modal purposes. It is applied to the castle of Jehay (Belgium), which is under renovation, to permit multi-user digital immersive experiences. The article proposes a high-level view of multi-disciplinary processes, from a needs analysis to the 3D reality capture workflow and the creation of a VR environment incorporated into an immersive application. We provide several relevant VR parameters for the scene optimization, the locomotion system, and the multi-user environment definition that were tested in a heritage tourism context [less ▲]

Detailed reference viewed: 52 (10 ULiège)
Full Text
See detailPoint Cloud vs. Mesh Features for Building Interior Classification
Bassier, Maarten; Vergauwen, Maarten; Poux, Florent ULiege

in Remote Sensing (2020), 12(14), 2224

Interpreting 3D point cloud data of the interior and exterior of buildings is essential for automated navigation, interaction and 3D reconstruction. However, the direct exploitation of the geometry is ... [more ▼]

Interpreting 3D point cloud data of the interior and exterior of buildings is essential for automated navigation, interaction and 3D reconstruction. However, the direct exploitation of the geometry is challenging due to inherent obstacles such as noise, occlusions, sparsity or variance in the density. Alternatively, 3D mesh geometries derived from point clouds benefit from preprocessing routines that can surmount these obstacles and potentially result in more refined geometry and topology descriptions. In this article, we provide a rigorous comparison of both geometries for scene interpretation. We present an empirical study on the suitability of both geometries for the feature extraction and classification. More specifically, we study the impact for the retrieval of structural building components in a realistic environment which is a major endeavor in Building Information Modeling (BIM) reconstruction. The study runs on segment-based structuration of both geometries and shows that both achieve recognition rates over 75% F1 score when suitable features are used. [less ▲]

Detailed reference viewed: 53 (9 ULiège)
Full Text
See detailFundamentals to clustering high-dimensional data (3D point clouds)
Poux, Florent ULiege

Article for general public (2020)

Clustering algorithms allow data to be partitioned into subgroups, or clusters, in an unsupervised manner. Intuitively, these segments group similar observations together. Clustering algorithms are ... [more ▼]

Clustering algorithms allow data to be partitioned into subgroups, or clusters, in an unsupervised manner. Intuitively, these segments group similar observations together. Clustering algorithms are therefore highly dependent on how one defines this notion of similarity, which is often specific to the field of application. Why unsupervised segmentation & clustering is the “bulk of AI”? What to look for when using them? How to evaluate performances? Explications and Illustration over 3D point cloud data. [less ▲]

Detailed reference viewed: 50 (1 ULiège)
Full Text
See detailAutomatic extraction and management of semantics within point cloud data
Poux, Florent ULiege

Conference (2020, May 28)

In the context of 3D point clouds, reality capture and artificial intelligence, the presentation covers exciting research potential of point clouds, semantics and unsupervised frameworks.

Detailed reference viewed: 110 (14 ULiège)
Full Text
See detailHow to represent 3D Data?
Poux, Florent ULiege

in Towards Data Science (2020, May 11)

The 3D datasets in our computerized ecosystem — of which an increasing number comes directly from reality capture devices — are found in different forms that vary in both the structure and the properties ... [more ▼]

The 3D datasets in our computerized ecosystem — of which an increasing number comes directly from reality capture devices — are found in different forms that vary in both the structure and the properties. Interestingly, they can be somehow mapped with success to point clouds thanks to its canonical nature. This article gives you the main 3D data representations modes to choose from when bindings point clouds to your application [less ▲]

Detailed reference viewed: 93 (14 ULiège)
Full Text
See detail5-Step Guide to generate 3D meshes from point clouds with Python
Poux, Florent ULiege

Article for general public (2020)

Tutorial to generate 3D meshes (.obj, .ply, .stl, .gltf) automatically from 3D point clouds using python. (Bonus) Surface reconstruction to create several Levels of Detail.

Detailed reference viewed: 34 (1 ULiège)
Full Text
See detailDiscover 3D Point Cloud Processing with Python
Poux, Florent ULiege

Article for general public (2020)

Tutorial to simply set up your python environment, start processing and visualize 3D point cloud data. Highlights Anaconda, NumPy, Matplotlib and Google Colab

Detailed reference viewed: 28 (1 ULiège)
Full Text
See detailThe Future of 3D Point Clouds: a new perspective
Poux, Florent ULiege

Article for general public (2020)

Discrete spatial datasets known as point clouds often lay the groundwork for decision-making applications. But can they become the next big thing?

Detailed reference viewed: 19 (1 ULiège)
Full Text
See detailThe Smart Point Cloud Model: Integration of point intelligence
Poux, Florent ULiege

Conference (2019, December 05)

Point cloud acquisition and processing workflows are usually application-dependent following a classic progression from data gathering to deliverable creation. While the collection step may be specific to ... [more ▼]

Point cloud acquisition and processing workflows are usually application-dependent following a classic progression from data gathering to deliverable creation. While the collection step may be specific to the sensor at hands, point-cloud-as-a-deliverable upsurges, becoming one de-facto choice for many industries. This task-oriented scenario mainly considers these as a spatial reference – which is used by experts to create other deliverables – thus being a project’s closest link to reality. It brings accurate real-world information which could allow decision-making based on digital-reality instead of interpreted or not up-to-date information. However, there are several considerations to address for a suitable integration. Point clouds are often very large depending on how much data is collected – usually in the realms of Gigabytes, if not Terabytes – and are destined to be archived as a reusable support to create new type of data and products. This can lead to a dead-end with exponential storage needs, incompatibility between outputs, loss of information and complicated collaboration. These practices also show limited to no attempt to generalize a framework which could in turn play as a common ground for further interoperability and generalization. This lack is counterproductive and could lead in term to a chaotic data repartition among actors and worsen the dependency to several outsourced service, each aiming an application independently. This primarily emphasize a strong need to study interoperable scenarios in which one point cloud could be used by many users from different domains, each having a different need (E.g. the object of interest can be a building or only the roof of this building). This will in turn introduce new constraints at the acquisition level to define the needed exhaustivity of the 3D representation for use with reasoning engines. Of course, this serialize additional challenges for interconnecting processes and insuring a compatibility with the different sources, volumes and other data-driven parameters. Secondly, robotics research has made a leap forward providing autonomous 3D recording systems, where we obtain a 3D point cloud of environments with no human intervention. Of course, following this idea to develop autonomous surveying demands that the data can be used for decision-making. The collected point cloud without context does not permit to take a valid decision, and the knowledge of experts is needed to extract the necessary information and to creates a viable data support for decision-making. Automating this process for fully autonomous cognitive decision systems is very tempting but poses many challenges mainly link to Knowledge Extraction (KE), Knowledge Integration (KI) and Knowledge Representation (KR) from point cloud. Therefore, point cloud structuration must be specifically designed to allow the computer to use it as a base for information extraction using reasoning and agent-based systems. Interoperable approaches which permits several actors to leverage one common information system (E.g. Facility Management 4.0) based on a digital twin is a great exploration motor. In this continuum, the presentation feeds a broader reflexion to go from a human-centered process to an autonomous workflow which highlights a need to improve automation, data management and interaction to speed-up inference processes, crucial to the development of point clouds in 3D capture workflows. The presentation primarily aims at providing all the necessary information for the development of an infrastructure: The Smart Point Cloud (SPC). It permits to handle point cloud data, manage heterogeneity, process and group points that retain a relationship regarding a specific domain ontology that allow to query and reason for decision-making tools including smart modelling. The resulting implementation of the SPC is based on new meta-models that permit to structure the information (3D geometry and semantics) and leverage available knowledge for accessing decision-making support tools and reasoning capabilities. At the frontier between a point cloud GIS system and a spatial infrastructure for agent-based decision support systems, its flexibility allows to evolve with future developments using artificial intelligence and new machine learning approaches. The proposed modular infrastructure includes Knowledge Discovery processes with Knowledge Integration and Knowledge Representation as ontologies, proving efficient context-specific adaptation. [less ▲]

Detailed reference viewed: 120 (13 ULiège)