Реконструкція 3D моделі об'єкта з використанням монокулярної RGB камери
Every year the edge between the real and digital worlds is becoming more and more blurred. Augmented and virtual reality rapid development creates new opportunities for more productive work and entertainment, revolution in 3D printing technologies begets boost in multiple DIY communities appearance...
Saved in:
| Date: | 2017 |
|---|---|
| Main Authors: | , , |
| Format: | Article |
| Language: | English |
| Published: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2017
|
| Subjects: | |
| Online Access: | https://journal.iasa.kpi.ua/article/view/119388 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Journal Title: | System research and information technologies |
| Download file: |  |
Institution
System research and information technologies| _version_ | 1867334327144022016 |
|---|---|
| author | Vedmedenko, Oleg V. Nikolaiev, Sergii S. Tymoshenko, Y. A. |
| author_facet | Vedmedenko, Oleg V. Nikolaiev, Sergii S. Tymoshenko, Y. A. |
| author_institution_txt_mv | [
{
"author": "Oleg V. Vedmedenko",
"institution": "Educational and Scientific Complex \"Institute for Applied System Analysis\" of the National Technical University of Ukraine \"Igor Sikorsky Kyiv Polytechnic Institute\", Kyiv"
},
{
"author": "Sergii S. Nikolaiev",
"institution": "Educational and Scientific Complex \"Institute for Applied System Analysis\" of the National Technical University of Ukraine \"Igor Sikorsky Kyiv Polytechnic Institute\", Kyiv"
},
{
"author": "Y. A. Tymoshenko",
"institution": "Educational and Scientific Complex \"Institute for Applied System Analysis\" of the National Technical University of Ukraine \"Igor Sikorsky Kyiv Polytechnic Institute\", Kyiv"
}
] |
| author_sort | Vedmedenko, Oleg V. |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2018-04-04T16:37:16Z |
| description | Every year the edge between the real and digital worlds is becoming more and more blurred. Augmented and virtual reality rapid development creates new opportunities for more productive work and entertainment, revolution in 3D printing technologies begets boost in multiple DIY communities appearance and sharing economy growth. All these factors require new technologies that allow making 3D models from real world objects, but most of these solutions are either very expensive or require complex technical knowledge that most ordinary people do not have. This paper provides a review and comparison of modern methods for 3D models of physical objects real time reconstruction that can be used in present-day mobile solutions. |
| doi_str_mv | 10.20535/SRIT.2308-8893.2017.4.02 |
| first_indexed | 2025-07-17T10:23:20Z |
| format | Article |
| fulltext |
O.V. Vedmedenko, S.S. Nikolaiev, Y.A. Tymoshenko, 2017
20 ISSN 1681–6048 System Research & Information Technologies, 2017, № 4
УДК 004.932.2
DOI: 10.20535/SRIT.2308-8893.2017.4.02
3D-MODEL RECONSTRUCTION
WITH USE OF MONOCULAR RGB CAMERA
O.V. VEDMEDENKO, S.S. NIKOLAIEV, Y.A. TYMOSHENKO
Abstract. Every year the edge between the real and digital worlds is becoming more
and more blurred. Augmented and virtual reality rapid development creates new op-
portunities for more productive work and entertainment, revolution in 3D printing
technologies begets boost in multiple DIY communities appearance and sharing
economy growth. All these factors require new technologies that allow making 3D
models from real world objects, but most of these solutions are either very expensive
or require complex technical knowledge that most ordinary people do not have. This
paper provides a review and comparison of modern methods for 3D models of phys-
ical objects real time reconstruction that can be used in present-day mobile solu-
tions.
Keywords: 3D model, 3D object, simultaneous localization and mapping problem,
SLAM, monocular camera, RGB camera, LSD-SLAM, ORB-SLAM.
INTRODUCTION
Nowadays a rapid transition from two-dimensional to three-dimensional informa-
tion space can be observed: flat interfaces are becoming less popular, three-
dimensional printing technology has achieved great success in a wide range of
different industries, virtual and augmented reality are no longer only ideas of sci-
ence fiction.
Each of the examples above requires a certain electronic volumetric repre-
sentation — a 3D-model. 3D interface requires a layout in space, 3D printing
technology requires a model of a 3D object, virtual and augmented reality are
both, in essence, just a manipulation and display of various 3D models.
However, the creation of three-dimensional content is a very difficult and
painstaking work. Designers and 3D illustrators spend a lot of resources to de-
velop even a simple model despite the fact that we see many of them in our daily
lives, and those we don’t see are often only modifications of objects from the real
world. We use the following methods directly to solve such problems. Those
methods provide a possibility to turn your smartphone into a monocular 3D scan-
ner — it’s very convenient to have a model, which is ready for further modifica-
tion and use of other software, just a few minutes after you saw it.
METHODS CLASSIFICATION
Today, having special knowledge, 3D model can be reconstructed with a number
of methods that solve the problem described in the literature as the problem of
simultaneous localization and mapping (i.e. SLAM). There are many different
implementations of this problem and they can be classified by:
3D-Model reconstruction with use of monocular RGB camera
Системні дослідження та інформаційні технології, 2017, № 4 21
1. The way algorithm analyze image information:
a. Ones which do not use features or key points (feature-less);
b. Ones which do use features (feature-based).
2. The type of sensor used to capture images: RGB camera, RGB-D camera
[1], laser range finders [2].
3. The number of cameras the video stream is received from:
a. Ones which use one camera (monocular);
b. Ones which use two or more cameras (stereo).
ORB-SLAM METHOD
ORB-SLAM [3] is a feature-based monocular SLAM system which can operate in
real time, in all kinds of environments. The system is robust to sharp and fast mo-
tion, allows fast loop closing algorithm and relocalization algorithm. Also it in-
cludes automatic initialization. It uses a survival of the fittest strategy that aimed
to select unique points and key frames, which leads to excellent robustness and
generates a compact map that grows only when environment varies in time.
Main design idea of ORB-SLAM system is that the same remembered key
features are used by the mapping, tracking, and for place recognition to perform
relocalization and loop closing. This allows to avoid the need to calculate the
depths of recognized features. ORB-SLAM system requires algorithm which can
extract features from image in less than 33ms per image, which excludes the
popularSIFT (~300 ms) [4], SURF (~300 ms) [5] or the recent A-KAZE (~100
ms) [6]. To obtain general place recognition capabilities, it also requires rotation
invariance, which excludes BRIEF [7] and LDB [8].
ORB-SLAM uses ORB [9] method. It is extremely fast, while it has good
invariance to viewpoint. This boosts the accuracy of bundle adjustment.
ORB-SLAM system incorporates three components which are running si-
multaneously: tracking component, local mapping component and loop closing
component.
The tracking component is responsible for localizing the camera with every
frame and deciding when to create a new key frame. Algorithm performs feature
comparison between current and previous frame and optimize the pose using mo-
tion-only bundle adjustment. If the tracking for any reason is lost (e.g. because of
occlusions or sharp movements), the place recognition module is responsible to
perform a global relocalization. When there is an initial estimation of the camera
position and feature matchings, a local visible map is reclaimed using the covisi-
bility graph consisting of key frames that is remembered by the system. Then
matches with the local map points are searched by reprojection, and camera posi-
tion is optimized again with all found matches. Finally the tracking thread is in-
serting a new key frame if needed.
All visual SLAM works in the literature agree that running bundle adjust-
ment with all the points and all the frames is not feasible. The work of Strasdat et
al. [10] showed that the most cost effective approach is to keep as much points as
possible, while keeping only non-redundant key frames. One approach is to insert
O.V. Vedmedenko, S.S. Nikolaiev, Y.A. Tymoshenko
ISSN 1681–6048 System Research & Information Technologies, 2017, № 4 22
key frames very cautiously to avoid an excessive growth of the computational
complexity. This restrictive key frame insertion algorithm makes the tracking fail
in hard exploration conditions. ORB-SLAM uses survival of the fittest strategy
which achieves unprecedented robustness in difficult scenarios by inserting key
frames as quickly as possible, and removing later the redundant ones, to avoid the
extra cost. This permits a flexible map expansion during exploration, which in-
creases tracking robustness in hard conditions (e.g. camera rotation, fast and sharp
movements), while its size is bounded in continual revisits to the same environ-
ment, i.e. lifelong operation.
The local mapping is responsible for processing new key frames added by
previous components and performs local bundle adjustment, which leads to opti-
mal reconstruction in areas, close to camera position. It is looking for a new fea-
tures in all new key frames by a comparison to connected key frames in the map,
and when it founds new feature it generates a new point. Based on the information
gathered during the work of the system, special algorithm is applied to cut all re-
dundant remembered points and save only most informative ones. Also it cuts all
redundant key frames.
Each key frame iK stores:
1. The camera pose iwT , which is a rigid body transformation that trans-
forms points from the world to the camera coordinate system.
2. The camera intrinsics, including focal length and principal point.
3. All the ORB features extracted in the frame, associated or not to a map
point, whose coordinates are undistorted if a distortion model is provided.
Each map point ip stores:
1. Its 3D position iwX , in the world coordinate system.
2. The viewing direction in , which is the mean unit vector of all its viewing
directions (the rays that join the point with the optical center of the key frames
that observe it).
3. A representative ORB descriptor iD , which is the associated ORB de-
scriptor whose hamming distance is minimum with respect to all other associated
descriptors in the key frames in which the point is observed.
4. The maximum maxd and minimum mind distances at which the point can
be observed, according to the scale invariance limits of the ORB features.
The loop closing searches for loops with every new key frame. If a loop is
detected, it computes a similarity transformation that informs about the drift ac-
cumulated in the loop. Then both sides of the loop are aligned and duplicated
points are merged. Finally algorithm performs a pose graph optimization over
similarity constraints to achieve global consistency.
Running ORB-SLAM system is demonstrated on fig. 1. At the bottom left
corner there is a current frame with features highlighted with green. At the right
side there is a built map: remembered features displayed with red dots, green
square is a current camera position and blue squares are key frames.
3D-Model reconstruction with use of monocular RGB camera
Системні дослідження та інформаційні технології, 2017, № 4 23
LSD-SLAM METHOD
Large-Scale Direct Monocular SLAM (LSD-SLAM) [11] — a direct (feature-
less) monocular SLAM algorithm. Along with highly accurate pose estimation
based on direct image alignment, it provides the 3D environment real-time recon-
struction as a graph of key frames with semi-dense depth maps. These are ob-
tained by a huge number of pixel-by-pixel comparisons.
The fundamental idea behind feature-based approaches is to split the overall
problem — estimating geometric information from images — into two sequential
steps: first, a set of feature observations is extracted from the image. Second, the
camera position and scene geometry is computed as a function of these feature
observations only. While this decoupling simplifies the overall problem, it comes
with an important limitation: only information that conforms to the feature type
can be used. In particular, when using keypoints, information contained in straight
or curved edges — which especially in man-made environments make up a large
part of the image — is discarded.
Direct visual odometry methods circumvent this limitation by optimizing the
geometry directly on the image intensities, which enables using all information in
the image. In addition to higher accuracy and robustness in particular in
environments with little keypoints, this provides substantially more information
about the geometry of the environment, which can be very valuable for robotics
or augmented reality applications.
A condensed summary of the relevant mathematical concepts is given in the
next paragraphs.
2Ω is the set of normalized pixel coordinates, i.e., they include the in-
trinsic camera calibration.
d is used to denote the inverse of the depth z of a point, i.e., 1 zd .
(3)SO is a rotations around a fixed point in three-dimensional Euclidean
space group, which consists of orthogonal 33 matrices with a determinant
equals to 1.
Fig 1. Running ORB-SLAM system
O.V. Vedmedenko, S.S. Nikolaiev, Y.A. Tymoshenko
ISSN 1681–6048 System Research & Information Technologies, 2017, № 4 24
A 3D rigid body transform (3)SEG denotes rotation and translation in
3D, i.e. is defined by
3and (3) with
10
tSOR
tR
G .
During optimization, a minimal representation for the camera pose is re-
quired, which is given by the corresponding element (3) SE of the associated
Lie-algebra. Elements are mapped to 3SE by the exponential map
)(exp (3) seG , its inverse being denoted by GSE 3log . With a slight
abuse of notation, element of (3)se can be used to represent pose and written as
vector 6R .
The transformation moving a point from frame i to frame j is written as
jiξ . For convenience, the pose concatenation operator (3)(3)(3): sesese
should be defined as
))(exp)((explog (3)(3)(3) jisekjseSEjikjki .
Further, 3D projective warp function ω should be defined, which projects an
image point p and its inverse depth d into a by ξ transformed camera frame
z
z
y
z
x
dp
1
),,( with
1
1
exp
1
(3)
d
d
p
d
p
z
y
x
y
x
se .
A 3D similarity transform (3) Sim S denotes rotation, scaling and transla-
tion, i.e. is defined by
stSOR
tsR
G and ,(3) with
10
3 .
As for rigid body transformations, a minimal representation is given by ele-
ments of the associated Lie-algebra (3) simξ , which now have an additional
degree of freedom, that is 7R . The exponential and logarithmic map, pose
concatenation and a projective warp function s can be defined analogously to
the (3)se case.
Propagation of uncertainty is a statistical tool to derive the uncertainty of the
output of a function )(Xf , caused by uncertainty on its input X . Assuming X
to be Gaussian distributed with covariance XΣ , the covariance of )(Xf can be
approximated (using the Jacobian fJ of f ) by
T
fXff JJ .
3D-Model reconstruction with use of monocular RGB camera
Системні дослідження та інформаційні технології, 2017, № 4 25
The algorithm consists of three major components: tracking, depth map es-
timation and map optimization.
The tracking component continuously tracks new camera images. That is, it
estimates their rigid body pose (3)se with respect to the current key frame,
using the pose of the previous frame as initialization.
The depth map estimation component uses tracked frames to either refine or
replace the current key frame. Depth is refined by filtering over many per-pixel,
small-baseline stereo comparisons coupled with interleaved spatial regularization.
If the camera moves too far away from the existing map, a new key frame is cre-
ated from the most recent tracked image. Algorithm threshold a weighted combi-
nation of relative distance and angle to the current key frame:
jijiji W T)(dist ,
where W is a diagonal matrix containing the weights.
Each key frame iK consists of a camera image iiI Ω: , an inverse
depth map
iDiD Ω: , and the variance of the inverse depth
ii DV Ω:
. Note that the depth map and variance are only defined for a subset
of pixels iDi
ΩΩ , containing all image regions in the vicinity of sufficiently
large intensity gradient, hence semi-dense. Edges jiE between key frames contain
their relative alignment as similarity transform (3)sim ji , as well as the corre-
sponding covariance matrix ji .
Once a key frame is replaced as tracking reference — and hence its depth
map will not be refined further — it is incorporated into the global map by the
map optimization component. To detect loop closures and scale-drift, a similarity
transform (3)sim to close-by existing key frames (including its direct prede-
cessor) is estimated using scale-aware, direct (3)sim — image alignment.
To bootstrap the LSD-SLAM system, it is sufficient to initialize a first key
frame with a random depth map and large variance. Given sufficient translational
camera movement in the first seconds, the algorithm “locks” to a certain configu-
ration, and after a couple of key frame propagations converges to a correct depth
configuration.
Starting from an existing key frame ),,( iiii VDIK , the relative 3D pose
(3)ξ seij of a new image jI is computed by minimizing the variance-
normalized photometric error
2
), (
2
), (
)(
jipiD pr
jip
p
jip
pr
E ,
with ))), (, (()(), ( jiijijip pDpIpIpr ,
)(
)(
), (
2
2
22
), ( pV
pD
pr
i
i
jip
Ipr jip
,
O.V. Vedmedenko, S.S. Nikolaiev, Y.A. Tymoshenko
ISSN 1681–6048 System Research & Information Technologies, 2017, № 4 26
where
δ
is the Huber norm applied to the normalized residual
.otherwise
2
, if,
2
2
r
r
r
r
The residual’s variance 2
), (σ
jip pr is computed using covariance propagation
as described above, utilizing the inverse depth variance iV and assuming Gaus-
sian image intensity noise 2
Iσ .
Running LSD-SLAM system is demonstrated on fig. 2. At the top left corner
there is a current frame captured by the camera. At the bottom left there is a cur-
rent key frame with color-coded depth map (from red — close objects, to blue —
far objects). At the right side there is a built point cloud with red square as a cur-
rent camera position and blue ones as camera trajectory.
The authors of this work were able to obtain promising results using the
LSD-SLAM method (see fig. 3–6).
Fig 2. Running LSD-SLAM system
Fig 3. Experiment 1. Object
3D-Model reconstruction with use of monocular RGB camera
Системні дослідження та інформаційні технології, 2017, № 4 27
METHODS COMPARISON
It’s convenient to compare those methods with a following table.
LSD-SLAM compared to ORB-SLAM
Method Trajectory loss Relocalization Point cloud Speed of work
LSD-SLAM
Method is unstable,
trajectory
is often lost
Slow and not
precise
Monochrome point
cloud from RGB
camera
Works in real
time, but re-
quires fast GPU
ORB-SLAM
Method is robust,
trajectory is lost
only for very sharp
movements in the
direction
of non-scanned areas
Fast
and precise
Colored point cloud
only from RGB-D
camera
Works in real
time, fast GPU
is not required
Fig. 5. Experiment 2. Object
Fig. 6. Experiment 2. Result
Fig. 4. Experiment 1. Result
O.V. Vedmedenko, S.S. Nikolaiev, Y.A. Tymoshenko
ISSN 1681–6048 System Research & Information Technologies, 2017, № 4 28
CONCLUSIONS
The analysis of these methods and their comparison give the following results. If
the task is to only localize camera position and build movement trajectory it is
clearly necessary to choose ORB-SLAM. In case the user can handle RGB-D
camera it is also possible to build colored point cloud representing the three-
dimensional model of the environment or object. Instead, if it is crucial to build a
3D model a user needs to choose LSD-SLAM even though the scans may have
worse quality.
REFERENCES
1. An evaluation of the RGB-D SLAM system / F. Endres, J. Hess, N. Engelhard etc. //
Robotics and Automation (ICRA). — 2012. // IEEE International Conference. —
P. 1691–1696.
2. Robot pose estimation in unknown environments by matching 2d range scans / Lu F.,
Milios E. // Journal of Intelligent and Robotic Systems. — 1997. — Vol. 18,
№ 3. — P. 249–275.
3. ORB-SLAM: A Versatile and Accurate Monocular SLAM System / Raúl Mur-Artal, J.
M. M. Montiel, Juan D. Tardós // IEEE Transactions on Robotics. — 2015. —
Vol. 31, N 5. — P. 1147–1163.
4. Distinctive image features from scale-invariant keypoints / D. G. Lowe // Interna-
tional Journal of Computer Vision. — 2004. — Vol. 60, N 2. — P. 91–110.
5. SURF: Speeded Up Robust Features / H. Bay, T. Tuytelaars, and L. Van Gool // Eu-
ropean Conference on Computer Vision (ECCV). — 2006. — P. 404–417.
6. Fast explicit diffusion for accelerated features in nonlinear scale spaces / P. F. Alcan-
tarilla, J. Nuevo, and A. Bartoli // British Machine Vision Conference (BMVC).
— 2013. — P. 1–11.
7. BRIEF: Binary Robust Independent Elementary Features / M. Calonder, V. Lepetit,
C. Strecha, and P. Fua // European Conference on Computer Vision (ECCV). —
2010. — P. 778–792.
8. LDB: An ultra-fast feature for scalable augmented reality on mobile devices /
X. Yang and K.-T. Cheng // IEEE International Symposium on Mixed and Aug-
mented Reality. — 2012. — P. 49–57.
9. ORB: an efficient alternative to SIFT or SURF / E. Rublee, V. Rabaud, K. Konolige,
G. Bradski // IEEE International Conference on Computer Vision (ICCV). —
2011. — P. 2564–2571.
10. Visual SLAM: Why filter? / H. Strasdat, J. M. M. Montiel, A. J. Davison // Image
and Vision Computing. — 2012. — Vol. 30, № 2. — P. 65–77.
11. LSD-SLAM: Large-Scale Direct Monocular SLAM / J. Engel, T. Schöps, D. Cremers
// European Conference on Computer Vision (ECCV). — 2014. — P. 834–849.
Received 23.06.2017
From the Editorial Board: the article corresponds completely to submitted manuscript.
|
| id | journaliasakpiua-article-119388 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2025-07-17T10:23:20Z |
| publishDate | 2017 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/6c/2b329ab0a905070fe9541d6bccf3576c.pdf |
| spelling | journaliasakpiua-article-1193882018-04-04T16:37:16Z 3D-model reconstruction with use of monocular RGB camera Реконструкция 3D модели объекта с использованием монокулярной RGB камеры Реконструкція 3D моделі об'єкта з використанням монокулярної RGB камери Vedmedenko, Oleg V. Nikolaiev, Sergii S. Tymoshenko, Y. A. 3D model 3D object simultaneous localization and mapping problem SLAM monocular camera RGB camera LSD-SLAM ORB-SLAM 3D модель 3D объект проблема одновременной локализации и построения карты SLAM монокулярная камера RGB камера LSD-SLAM ORB-SLAM 3D модель 3D об’єкт проблема одночасної локалізації та побудови карти SLAM монокулярна камера RGB камера LSD-SLAM ORB-SLAM Every year the edge between the real and digital worlds is becoming more and more blurred. Augmented and virtual reality rapid development creates new opportunities for more productive work and entertainment, revolution in 3D printing technologies begets boost in multiple DIY communities appearance and sharing economy growth. All these factors require new technologies that allow making 3D models from real world objects, but most of these solutions are either very expensive or require complex technical knowledge that most ordinary people do not have. This paper provides a review and comparison of modern methods for 3D models of physical objects real time reconstruction that can be used in present-day mobile solutions. Ежегодно границы реального и цифрового мира становятся все более размытыми. Быстрое развитие дополненной и виртуальной реальности создает новые возможности для более продуктивной работы и развлечений; революция в технологиях 3D-печати порождает появление множества DIY сообществ и способствует росту экономики. Все эти факторы требуют новых технологий, которые позволят создавать 3D-модели из объектов реального мира, а большинство из доступных решений либо очень дорогостоящие, либо требуют сложных технических знаний, которых у большинства людей нет. Представлено обзор и сравнены современные методы реконструкции 3D моделей реальных объектов в реальном времени, которые могут быть использованы в современных мобильных решениях. Щорічно межі між реальним і цифровим світом стають дедалі більш розмитими. Швидкий розвиток доповненої і віртуальної реальності створює нові можливості для більш продуктивної роботи і розваг; революція в технологіях 3D-друку породжує появу безлічі DIY спільнот і сприяє зростанню економіки. Усі ці фактори потребують нових технологій, які дозволять створювати 3D-моделі з об'єктів реа-льного світу, а більшість з доступних рішень або дуже дорогі, або потребують складних технічних знань, яких у більшості людей немає. Подано огляд і порівняно сучасні методи реконструкції 3D моделей реальних об’єктів у реальному часі, які можуть бути використані в сучасних мобільних рішеннях. The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2017-12-15 Article Article application/pdf https://journal.iasa.kpi.ua/article/view/119388 10.20535/SRIT.2308-8893.2017.4.02 System research and information technologies; No. 4 (2017); 20-28 Системные исследования и информационные технологии; № 4 (2017); 20-28 Системні дослідження та інформаційні технології; № 4 (2017); 20-28 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/119388/114229 Copyright (c) 2021 System research and information technologies |
| spellingShingle | 3D модель 3D об’єкт проблема одночасної локалізації та побудови карти SLAM монокулярна камера RGB камера LSD-SLAM ORB-SLAM Vedmedenko, Oleg V. Nikolaiev, Sergii S. Tymoshenko, Y. A. Реконструкція 3D моделі об'єкта з використанням монокулярної RGB камери |
| title | Реконструкція 3D моделі об'єкта з використанням монокулярної RGB камери |
| title_alt | 3D-model reconstruction with use of monocular RGB camera Реконструкция 3D модели объекта с использованием монокулярной RGB камеры |
| title_full | Реконструкція 3D моделі об'єкта з використанням монокулярної RGB камери |
| title_fullStr | Реконструкція 3D моделі об'єкта з використанням монокулярної RGB камери |
| title_full_unstemmed | Реконструкція 3D моделі об'єкта з використанням монокулярної RGB камери |
| title_short | Реконструкція 3D моделі об'єкта з використанням монокулярної RGB камери |
| title_sort | реконструкція 3d моделі об'єкта з використанням монокулярної rgb камери |
| topic | 3D модель 3D об’єкт проблема одночасної локалізації та побудови карти SLAM монокулярна камера RGB камера LSD-SLAM ORB-SLAM |
| topic_facet | 3D model 3D object simultaneous localization and mapping problem SLAM monocular camera RGB camera LSD-SLAM ORB-SLAM 3D модель 3D объект проблема одновременной локализации и построения карты SLAM монокулярная камера RGB камера LSD-SLAM ORB-SLAM 3D модель 3D об’єкт проблема одночасної локалізації та побудови карти SLAM монокулярна камера RGB камера LSD-SLAM ORB-SLAM |
| url | https://journal.iasa.kpi.ua/article/view/119388 |
| work_keys_str_mv | AT vedmedenkoolegv 3dmodelreconstructionwithuseofmonocularrgbcamera AT nikolaievsergiis 3dmodelreconstructionwithuseofmonocularrgbcamera AT tymoshenkoya 3dmodelreconstructionwithuseofmonocularrgbcamera AT vedmedenkoolegv rekonstrukciâ3dmodeliobʺektasispolʹzovaniemmonokulârnojrgbkamery AT nikolaievsergiis rekonstrukciâ3dmodeliobʺektasispolʹzovaniemmonokulârnojrgbkamery AT tymoshenkoya rekonstrukciâ3dmodeliobʺektasispolʹzovaniemmonokulârnojrgbkamery AT vedmedenkoolegv rekonstrukcíâ3dmodelíobêktazvikoristannâmmonokulârnoírgbkameri AT nikolaievsergiis rekonstrukcíâ3dmodelíobêktazvikoristannâmmonokulârnoírgbkameri AT tymoshenkoya rekonstrukcíâ3dmodelíobêktazvikoristannâmmonokulârnoírgbkameri |