Filters
cris logo
-
New search Filters
Select from list
Displayed from Show more
You have reached the maximum number of displayed search results.
All results displayed
No results are available for the search performed
Displayed from
You have reached the maximum number of displayed search results.
All results displayed
Loading results
Obtaining statistical data

Projects / Programmes source: ARIS

A neural network solution to segmentation and recovery of superquadric models from 3D image data

Edit
Research activity

Code Science Field Subfield
2.07.00  Engineering sciences and technologies  Computer science and informatics   

Code Science Field
P170  Natural sciences and mathematics  Computer science, numerical analysis, systems, control 

Code Science Field
1.02  Natural Sciences  Computer and information sciences 
Keywords
deep neural networks, 3D data, volumetric models, segmentation and reconstruction of geometric models
Evaluation (rules)
source: COBISS
Evaluation of bibliographic research performance indicators according to ARIS methodology
Citations Citations for bibliographic records in COBIB.SI that are linked to records in citation databases
source: WoS source: Scopus source: COBISS
Researchers (20)
no. Code Name and surname Research area Role Period No. of publications
1.  22472  PhD Borut Batagelj  Computer science and informatics  Researcher  2018 - 2020  192 
2.  53867  Blaž Bortolato  Physics  Researcher  2019  12 
3.  31252  PhD Narvika Bovcon  Computer science and informatics  Researcher  2019  308 
4.  54426  Tea Brašanac  Computer science and informatics  Researcher  2020  0 
5.  11805  PhD Simon Dobrišek  Computer science and informatics  Researcher  2018 - 2022  284 
6.  32737  PhD Niko Gamulin  Telecommunications  Researcher  2019 - 2020  8 
7.  38118  PhD Klemen Grm  Systems and cybernetics  Researcher  2018 - 2022  45 
8.  11161  PhD Aleš Jaklič  Computer science and informatics  Researcher  2018 - 2020  119 
9.  32887  MSc Bojan Klemenc  Computer science and informatics  Researcher  2019  56 
10.  31985  PhD Janez Križaj  Systems and cybernetics  Researcher  2018 - 2022  39 
11.  52331  Ivan Majhen    Technical associate  2019  0 
12.  53819  Blaž Meden  Computer science and informatics  Researcher  2019  56 
13.  54781  Tim Oblak  Computer science and informatics  Researcher  2020  14 
14.  36688  Klemen Pečnik  Telecommunications  Researcher  2019  34 
15.  19226  PhD Peter Peer  Computer science and informatics  Researcher  2018 - 2020  408 
16.  09581  PhD Franc Solina  Computer science and informatics  Head  2018 - 2020  640 
17.  23347  PhD Gregor Strle  Computer science and informatics  Researcher  2019  67 
18.  23401  PhD Luka Šajn  Computer science and informatics  Researcher  2019  108 
19.  53774  Jaka Šircelj  Computer science and informatics  Researcher  2019 - 2022  7 
20.  28458  PhD Vitomir Štruc  Systems and cybernetics  Researcher  2018 - 2022  361 
Organisations (2)
no. Code Research organisation City Registration number No. of publications
1.  1539  University of Ljubljana, Faculty of Computer and Information Science  Ljubljana  1627023  16,242 
2.  1538  University of Ljubljana, Faculty of Electrical Engineering  Ljubljana  1626965  27,771 
Abstract
Computer vision tries to replicate, at least partially, the functionality of human visual perception. Some of the many goals of visual perception is to enable our interaction with the physical world which is surrounding us, such as moving around without bumping into obstacles, grasping and touching of objects, and recognizing objects on several levels of abstraction. It has been acknowledged quite early in the progress of computer vision that to achieve these goals, the visual information must be at some point represented in terms of spatial or volumetric models since they can be directly related to the actual 3D physical space that surrounds us. One of the still popular volumetric part-level models where the actual 3D shape needs to be represented are superquadrics. Superquadrics are defined by a closed surface that can take up the shape of ellipsoids, cylinders, parallelopipeds, and all shapes in-between. They are popular in robotics, for example for grasp planning of previously unknown objects.  We developed in the 1990s the state of the art method for segmentation and reconstruction of superquadrics from range images. The method is still popular and quite widely used which is testified by many citations in Google Scholar (1500 citations anytime, 100 citations since 2014). There were two reasons that prevented a wider use of this modeling approach in the past: lack or a high cost of acquiring 3D data iterative method of model recovery that made the method not suitable for real-time applications. Due to the hardware and algorithmic advances in the past decade there is now a multitude of new methods and devices to acquire 3D image data. However, the iterative nature of the original superquadric recovery method still prevents its use when real-time operation is required. The path to a faster method is actually quite evident—use deep neural networks which have revolutionized computer vision research in the past few years. During the last few years, Convolutional Neural Networks (CNN) are slowly but surely becoming the default method solve many computer vision related problems. CNN based computational approach in computer vision in general is very fast, can cope with large data input, and has also similarities with the way how our brains are coping with processing of visual data.  We propose therefore in this project proposal to implement segmentation and superquadric model recovery using CNNs. As input to CNNs not only range images should be considered, but 3D point clouds in general. There are two types of applications that would benefit greatly from the results of this project: applications where real-time operation is required, such as in autonomous driving, applications where huge amounts of 3D data is generated (LiDAR, multi-image photogrammetry) and some intelligent automated processing of such data is needed. In the proposed project group we have ample experience with superquadric modeling since we are the authors of the state of the art method. On the other hand, we have also ample experience in developing CNN solutions for computer vision tasks. This makes us exceptionally qualified for the proposed project.
Significance for science
The objective of this research proposal is to develop a CNN solution for real-time segmentation and superquadric model recovery from large 3D point clouds. In addition to the development of CNNs for segmentation and model recovery of superquadrics from 3D point clouds, we would like to find out if these CNNs for segmentation and model recovery from 3D point clouds could be adapted to reconstruction from 2D intensity images. There is ample evidence by current research that the marriage of 3D data and models with CNN computational paradigm is adequate but only starting. Our motivation, however, is to develop a general purpose CNN based solution which can give for a given selected scene, defined with corresponding 3D point clouds and/or intensity images, its description in terms of supequadrics as part-level models. The output of our proposed solution would therefore be the parameter values of an unspecified number of superquadrics, which are necessary to describe a given scene. To our knowledge, no method exists yet for recovery of part-level volumetric models, such as superquadrics from 3D point clouds using CNNs. This research project would contribute to the growing field of 3D recovery and modeling using CNNs.  Since no other method exists that would be as fast as using deep neural networks for recovery of volumetric part-based models, the success of the proposed research would have a huge impact in application areas where real-time processing is required and when huge sets of 3D data points need to be interpreted. These application areas where real-time processing is needed are primarily robotics in unconstrained environments, where previously unknown objects can be encountered and must be modeled, such as in autonomous driving, handling of different objects, path planning, etc. Knowledge-based interpretation of huge sets of 3D data points obtained by LiDAR and multi-image photogrammetry could be achieved with faster methods.
Significance for the country
The objective of this research proposal is to develop a CNN solution for real-time segmentation and superquadric model recovery from large 3D point clouds. In addition to the development of CNNs for segmentation and model recovery of superquadrics from 3D point clouds, we would like to find out if these CNNs for segmentation and model recovery from 3D point clouds could be adapted to reconstruction from 2D intensity images. There is ample evidence by current research that the marriage of 3D data and models with CNN computational paradigm is adequate but only starting. Our motivation, however, is to develop a general purpose CNN based solution which can give for a given selected scene, defined with corresponding 3D point clouds and/or intensity images, its description in terms of supequadrics as part-level models. The output of our proposed solution would therefore be the parameter values of an unspecified number of superquadrics, which are necessary to describe a given scene. To our knowledge, no method exists yet for recovery of part-level volumetric models, such as superquadrics from 3D point clouds using CNNs. This research project would contribute to the growing field of 3D recovery and modeling using CNNs.  Since no other method exists that would be as fast as using deep neural networks for recovery of volumetric part-based models, the success of the proposed research would have a huge impact in application areas where real-time processing is required and when huge sets of 3D data points need to be interpreted. These application areas where real-time processing is needed are primarily robotics in unconstrained environments, where previously unknown objects can be encountered and must be modeled, such as in autonomous driving, handling of different objects, path planning, etc. Knowledge-based interpretation of huge sets of 3D data points obtained by LiDAR and multi-image photogrammetry could be achieved with faster methods.
Most important scientific results Interim report
Most important socioeconomically and culturally relevant results
Confirmation required
Views history
Favourite