Article
Computer Science, Artificial Intelligence
Suryansh Kumar, Yuchao Dai, Hongdong Li
Summary: This work proposes a unified approach to solve the task of dense 3D reconstruction of a complex dynamic scene from images, by modeling the dynamic scene as a combination of piecewise planar surfaces and local rigid motion, effectively simplifying the task and achieving state-of-the-art performance.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Computer Science, Information Systems
Cheng Feng, Long Ma, Congxuan Zhang, Zhen Chen, Liyue Ge, Shaofeng Jiang
Summary: In this article, a piecewise 3D scene flow estimation approach named SS-SF is proposed to address edge-blurring and enhance accuracy and robustness under motion occlusions. By utilizing semantic segmentation, optimizing mappings, and handling occlusions, the proposed method demonstrates advanced capabilities in scene flow estimation, especially in edge preservation and occlusion handling.
Article
Engineering, Civil
Cansen Jiang, Danda Pani Paudel, David Fofi, Yohan Fougerolle, Cedric Demonceaux
Summary: In this study, a framework is proposed to detect and extract moving objects from a sequence of unordered point clouds for building high-quality static maps. The accurate detection and segmentation of moving objects are achieved through 3D flow field analysis and sparse flow clustering algorithm.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Luis Roldao, Raoul de Charette, Anne Verroust-Blondet
Summary: This paper surveys the progress of semantic scene completion (SSC), highlighting the unresolved challenges and evaluating the performance of state-of-the-art techniques on popular datasets.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2022)
Article
Computer Science, Artificial Intelligence
Francisco Barranco, Cornelia Fermueller, Yiannis Aloimonos, Eduardo Ros
Summary: This paper investigates the impact of avoiding optical flow estimation on structure recovery, and proposes a new method based on image gradients to solve 3D motion problems by reformulating the positive-depth constraint. Experimental results show that the method achieves higher accuracy and outperforms existing techniques based on normal flow for 3D motion estimation.
PATTERN RECOGNITION
(2021)
Article
Computer Science, Theory & Methods
Thiago L. T. da Silveira, Paulo G. L. Pinto, Jeffri Murrugarra-Llerena, Claudio R. Jung
Summary: This article provides a comprehensive survey on pioneer and state-of-the-art 3D scene geometry estimation methodologies based on single, two, or multiple images captured under omnidirectional optics. It covers the basic concepts of spherical camera model, acquisition technologies, and representation formats. It also examines monocular, stereo, and multiple view camera setups, along with commonly adopted datasets and future trends.
ACM COMPUTING SURVEYS
(2023)
Article
Computer Science, Artificial Intelligence
Armin Mustafa, Marco Volino, Hansung Kim, Jean-Yves Guillemaut, Adrian Hilton
Summary: This paper introduces a general approach for obtaining a 4D representation of complex dynamic scenes from multi-view wide-baseline static or moving cameras without prior knowledge of the scene structure, appearance, or illumination. The contributions include automatic initial reconstruction, temporal coherence integration, and shape constraint, leading to improved accuracy in multi-view segmentation and dense reconstruction.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2021)
Article
Computer Science, Artificial Intelligence
Pia Bideau, Erik Learned-Miller, Cordelia Schmid, Karteek Alahari
Summary: A good understanding of geometry and familiarity with objects contribute to the reliable perception of moving objects. Human vision and computer vision differ in their approaches to this problem, with human vision coupling cognitive processes and body design, while computer vision relies on deep networks. The coupling of camera rotation and translation creates complex motion fields that are challenging for deep networks to untangle directly. This study presents a probabilistic model to estimate camera rotation and rectify the flow field for improved motion segmentation, yielding better results on benchmark tests.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2023)
Article
Engineering, Marine
Aurelien Arnaubec, Maxime Ferrera, Javier Escartin, Marjolaine Matabos, Nuno Gracias, Jan Opderbecke
Summary: This paper introduces an open source software, Matisse, for creating textured 3D models of complex underwater scenes from video or still images, addressing the lack of push-button software in optical marine imaging. Matisse allows for seamless integration of vehicle navigation data to produce georeferenced and properly scaled 3D reconstructions. The software also includes pre-processing tools and a 3D scene analysis tool, 3DMetrics, for extracting quantitative measurements from the 3D data analysis.
JOURNAL OF MARINE SCIENCE AND ENGINEERING
(2023)
Article
Mathematics
Mingyun Wen, Kyungeun Cho
Summary: Recent studies have shown that deep learning achieves excellent performance in reconstructing 3D scenes from multiview images or videos. However, these reconstructions do not provide the identities of objects, and object identification is necessary for a scene to be functional in virtual reality or interactive applications. To address this challenge, a module for depth feature generation and a novel approach to mesh reconstruction combining two decoders were proposed. The results demonstrated that the proposed method significantly improved the object-aware 3D scene reconstruction performance.
Article
Automation & Control Systems
Guangming Wang, Chaokang Jiang, Zehang Shen, Yanzi Miao, Hesheng Wang
Summary: The study utilizes generative adversarial networks to self-learn 3D scene flow and discriminates between real and synthesized point clouds, achieving accurate estimation of scene flow.
ADVANCED INTELLIGENT SYSTEMS
(2022)
Article
Optics
Di Liu, Jianfeng Sun, Wei Lu, Sining Li, Xin Zhou
Summary: In this paper, an algorithm for reconstructing dynamic scenes in Geiger-mode avalanche photodiode (GM-APD) light detection and ranging (LiDAR) is proposed. By extracting motion features and correcting data, the algorithm achieves super-resolution reconstruction and improves the signal-to-noise ratio and distance measurement compared to conventional algorithms. It also enables target detection, motion feature extraction, and position prediction, expanding the application scope of GM-APD LiDAR.
OPTICS AND LASER TECHNOLOGY
(2023)
Article
Engineering, Biomedical
Agniva Sengupta, Adrien Bartoli
Summary: The study aims to extract precise geometric information of the colonic surface from colonoscopic images by reconstructing the 3D surface of the colon. By developing a technique to compute the depth of feature points and fitting a deformable cylindrical model, the proposed method improves existing NRSfM methods and successfully obtains 3D reconstruction from real colonoscopic data.
INTERNATIONAL JOURNAL OF COMPUTER ASSISTED RADIOLOGY AND SURGERY
(2021)
Article
Computer Science, Artificial Intelligence
Wei Yin, Jianming Zhang, Oliver Wang, Simon Niklaus, Simon Chen, Yifan Liu, Chunhua Shen
Summary: Despite progress, challenges in depth estimation from single monocular images remain due to limited training data. To overcome this, a two-stage framework is proposed to predict depth and estimate scene shapes using large-scale relative depth data. Separate training of modules allows for flexibility, and image-level regression loss and normal-based geometry loss improve training with relative depth annotation. State-of-the-art performance is achieved on unseen datasets. Code available at: https://github.com/aim-uofa/depth/.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Johanna Wald, Nassir Navab, Federico Tombari
Summary: A 3D scene is not only about geometry and object classes, but also about the semantic network of interconnected nodes. While scene graphs have been proven effective in image tasks, we propose a new neural network architecture to learn semantic graphs from 3D scenes. Our method goes beyond object-level perception and explores relations between object entities.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2022)