Без темы
<<  Introduction to Mutual Funds Basic Portfolio Mathematics iPad Basic Training  >>
Introduction: Robot Vision Philippe Martinet Unifying Vision and
Introduction: Robot Vision Philippe Martinet Unifying Vision and
Wearable personal assistants
Wearable personal assistants
State representation
State representation
SLAM with first order uncertainty representation
SLAM with first order uncertainty representation
Introduction: Robot Vision Philippe Martinet Unifying Vision and
Introduction: Robot Vision Philippe Martinet Unifying Vision and
For data association, earlier approach
For data association, earlier approach
For data association, earlier approach
For data association, earlier approach
For data association, earlier approach
For data association, earlier approach
SIFT [D
SIFT [D
SIFT [D
SIFT [D
SIFT [D
SIFT [D
Uses SIFT-like descriptors (histogram of gradients) around Harris
Uses SIFT-like descriptors (histogram of gradients) around Harris
[Chekhlov, Pupilli, Mayol and Calway, ISVC06/CVPR07]
[Chekhlov, Pupilli, Mayol and Calway, ISVC06/CVPR07]
[Eade and Drummond, BMVC2006]
[Eade and Drummond, BMVC2006]
[Eade and Drummond, BMVC2006]
[Eade and Drummond, BMVC2006]
[Williams, Smith and Reid ICRA2007]
[Williams, Smith and Reid ICRA2007]
[Williams, Smith and Reid ICRA2007]
[Williams, Smith and Reid ICRA2007]
[Williams, Smith and Reid ICRA2007]
[Williams, Smith and Reid ICRA2007]
Relocalisation based on appearance hashing
Relocalisation based on appearance hashing
Parallel Tracking and Mapping
Parallel Tracking and Mapping
Discovering structure within SLAM
Discovering structure within SLAM
Discovering planes in SLAM
Discovering planes in SLAM
Discovering planes in SLAM
Discovering planes in SLAM
Mean error & State reduction, planes
Mean error & State reduction, planes
Discovering 3D lines
Discovering 3D lines
An example application
An example application
Картинки из презентации «Introduction: Robot Vision Philippe Martinet Unifying Vision and Control Selim Benhimane Efficient Keypoint Recognition Vincent Lepetit Multi-camera and Model-based Robot Vision Andrew Comport Visual SLAM for Spatially Aware Robots Walterio Mayol-Cuevas» к уроку информатики на тему «Без темы»

Автор: . Чтобы познакомиться с картинкой полного размера, нажмите на её эскиз. Чтобы можно было использовать все картинки для урока информатики, скачайте бесплатно презентацию «Introduction: Robot Vision Philippe Martinet Unifying Vision and Control Selim Benhimane Efficient Keypoint Recognition Vincent Lepetit Multi-camera and Model-based Robot Vision Andrew Comport Visual SLAM for Spatially Aware Robots Walterio Mayol-Cuevas.ppt» со всеми картинками в zip-архиве размером 2968 КБ.

Introduction: Robot Vision Philippe Martinet Unifying Vision and Control Selim Benhimane Efficient Keypoint Recognition Vincent Lepetit Multi-camera and Model-based Robot Vision Andrew Comport Visual SLAM for Spatially Aware Robots Walterio Mayol-Cuevas

содержание презентации «Introduction: Robot Vision Philippe Martinet Unifying Vision and Control Selim Benhimane Efficient Keypoint Recognition Vincent Lepetit Multi-camera and Model-based Robot Vision Andrew Comport Visual SLAM for Spatially Aware Robots Walterio Mayol-Cuevas.ppt»
Сл Текст Сл Текст
1Introduction: Robot Vision Philippe 21http://www.robots.ox.ac.uk/ActiveVision/Pr
Martinet Unifying Vision and Control Selim jects/Vslam/vslam.04/Videos/relocalisation
Benhimane Efficient Keypoint Recognition icra_07.mpg.
Vincent Lepetit Multi-camera and 22Relocalisation based on appearance
Model-based Robot Vision Andrew Comport hashing. Use a hash function to index
Visual SLAM for Spatially Aware Robots similar descriptors (Brown et al 2005).
Walterio Mayol-Cuevas Outdoor Visual SLAM Fast and memory efficient (only an index
for Robotics Kurt Konolige Advanced Vision needs to be saved per descriptor).
in Deformable Environments Adrien Bartoli. Quantize result of Haar masks. Chekhlov et
Tutorial organized by Andrew Comport and al 2008. Video at:
Adrien Bartoli Nice, September 22. http://www.cs.bris.ac.uk/Publications/pub_
2Visual SLAM and Spatial Awareness. aster.jsp?id=2000939.
SLAM= Simultaneous Localisation and 23Parallel Tracking and Mapping. [Klein
Mapping An overview of some methods and Murray, Parallel Tracking and Mapping
currently used for SLAM using computer for Small AR Workspaces Proc.
vision. Recent work on enabling more International Symposium on Mixed and
stable and/or robust mapping in real-time. Augmented Reality. 2007] Decouple Mapping
Work aiming to provide better scene from Tracking, run them in separate
understanding in the context of SLAM: threads on multi-core CPU. Mapping is
Spatial Awareness. Here we concentrate on based on key-frames, processed using batch
“Small” working areas where GPS, odometry Bundle Adjustment. Map is intialised from
and other traditional sensors are not a stereo pair (using 5-Point Algorithm).
operational or available. Initialised new points with epipolar
3Spatial Awareness. SA: A key cognitive search. Large numbers (thousands) of
competence that permits efficient motion points can be mapped in a small workspace.
and task planning. Even from early age we 24Parallel Tracking and Mapping. … CPU1.
use spatial awareness: the toy has not CPU2. [Klein and Murray, 2007]. Video at
vanished it is behind the sofa. I can http://www.robots.ox.ac.uk/ActiveVision/Vi
point to where the entrance to the eos/index.html.
building is but cant tell how many doors 25So far we have mentioned that. Maps
are from here to there. SLAM offers a are sparse collections of low-level
rigorous way to implement and manage SA. features: Points (Davison et al., Chekhlov
4Wearable personal assistants. Mayol, et al.) Edgelets (Eade and Drummond) Lines
Davison and Murray 2003. Video at (Smith et al., Gee and Mayol-Cuevas) Full
http://www.robots.ox.ac.uk/ActiveVision/Pr correlation between features and camera
jects/Vslam/vslam.02/Videos/wearableslam2. Maintain full covariance matrix Loop
pg. closure: effects of measurements
5SLAM. Key historical reference: Smith, propagated to all features in map Increase
R.C.and Cheeseman, P. "On the in state size limits number of features.
Representation and Estimation of Spatial 26Commonly in Visual SLAM. Emphasis on
Uncertainty". The International localization and less on the mapping
Journal of Robotics Research 5 (4): 56-68. output. SLAM should avoid making
1986. Proposed a stochastic framework to “beautiful” maps (there are other better
maintain the relationship (uncertainties) methods for that!). Very few examples
between features in the map. “Our exist on improving the awareness element,
knowledge of the spatial relationships e.g. Castle and Murray BMVC 07 on known
among objects is inherently uncertain. A object recognition within SLAM.
manmade object does not match its 27Better spatial awareness through
geometric model exactly because of higher level structural inference. Types
manufacturing tolerances. Even if it did, of Structure Coplanar points ? planes
a sensor could not measure the geometric Collinear edgelets ? lines Intersecting
features, and thus locate the object lines ? junctions Our Contribution Method
exactly, because of measurement errors. for augmenting SLAM map with planar and
And even if it could, a robot using the line structures. Evaluation of method in
sensor cannot manipulate the object simulated scene: discover trade-off
exactly as intended, because of hand between efficiency and accuracy.
positioning errors…”[Smith,Self,Cheesman 28Discovering structure within SLAM.
1986]. Gee, Checkhlov, Calway and Mayol-Cuevas,
6SLAM. A problem that has been 2008.
identified for several years, central in 29Plane Representation. Plane
mobile robot navigation and branching into Parameters: Basis vectors: Camera. normal.
other fields like wearable computing and (x,y,z). c(?2,?2). c(?1,?1). Plane. Gee et
augmented reality. al 2007.
7SLAM – Simultaneous Localisation And 30Plane Initialisation. O. Discover
Mapping. Aim to: Localise camera (6DOF – planes using RANSAC over thresholded
Rotation and Translation from reference subset of map Initialise plane in state
view) Simultaneously estimate 3D map of using best-fit plane parameters found from
features (e.g. 3D points). 3D points SVD of inliers Augment state covariance,
(features). update positions. perspective P, with new plane. P=. Gee et al 2007.
projection. update location. predict Append measurement covariance R0 to
location. Implemented using: Extended covariance matrix. Multiplication with
Kalman Filter, Particle filters, SIFT, Jacobian populates cross-covariance terms.
Edglets, etc. camera moved. camera. State size increases by 7 after adding
8State representation. as in [Davison plane.
2003]. 31Adding Points to Plane. ?max. d. s. O.
9SLAM with first order uncertainty Decide whether point lies on plane Add
representation. as in [Davison 2003]. point by projecting onto plane and
10 transforming state and covariance Decide
11Challenges for visual SLAM. On the whether to fix point on plane. Gee et al
computer vision side, improving data 2007. State size decreases by 1 after
association: Ensuring a match is a true adding point to plane. Add point to plane.
positive. Representations and Add other points to plane. Fix points in
parameterizations to enhance mapping while plane: reduces state size by 2 for each
within real-time. Alternative frameworks fixed point. State size is smaller than
for mapping: Can we extend area of original state if >7 points are added
operation? Better scene understanding. to plane.
12For data association, earlier 32Plane Observation. Cannot make direct
approach. Small (e.g. 11x11) image patches observation of plane Transform points to
around salient points to represent 3D world space Project points into image
features. Normalized Cross Correlation and match with predicted observations
(NCC) to detect features. Small patches + Covariance matrix embodies constraints
accurate search regions lead to fast between plane, camera and points. Gee et
camera pose estimation. Depth by al 2007.
projecting hypothesis at different depths. 33Discovering planes in SLAM. Gee et al.
See: A. Davison, Real-Time Simultaneous 2007. Video at:
Localisation and Mapping with a Single http://www.cs.bris.ac.uk/~gee.
Camera, ICCV 2003. 34Discovering planes in SLAM. Gee et al.
13However. Simple patches are 2007. Video at:
insufficient for large view point or scale http://www.cs.bris.ac.uk/~gee.
variations. Small patches help speed but 35Mean error & State reduction,
prone to mismatch. Search regions can’t planes. Average 30 runs. Gee at al 2008.
always be trusted (camera occlusion, 36Discovering 3D lines. Video at:
motion blur). Possible solutions: Use http://www.cs.bris.ac.uk/~gee.
better feature description or Other types 37An example application. Chekhlov et
of features e.g. edge information. al. 2007. Video at
14SIFT [D. Lowe, IJCV 2004]. Find maxima http://www.cs.bris.ac.uk/Publications/pub_
in scale space to locate keypoint. Around aster.jsp?id=2000745.
keypoint, build invariant local descriptor 38Other interesting recent work. Active
using gradient histograms. search and matching: or know what to
15Uses SIFT-like descriptors (histogram measure. Davison ICCV 2005 and Chli and
of gradients) around Harris corners. Get Davison ECCV 2008 Submapping: managing
scale from SLAM = “predictive SIFT”. better the scalability problem. Clemente
[Chekhlov, Pupilli, Mayol and Calway, et al RSS 2007 Eade and Drummond BMVC 2008
ISVC06/CVPR07]. And the work presented in this tutorial:
16[Chekhlov, Pupilli, Mayol and Calway, Randomised trees: Vincent Lepetit SFM:
ISVC06/CVPR07]. Video at Andrew Comport.
http://www.cs.bris.ac.uk/Publications/atta 39Software tools:
hment-delivery.jsp?id=9. http://www.doc.ic.ac.uk/~ajd/Scene/index.h
17[Eade and Drummond, BMVC2006]. ml <MonoSLAM code for Linux, works out
Edglets: Locally straight section of of the box>
gradient Image. Parameterized as 3D point http://www.robots.ox.ac.uk/~gk/PTAM/
+ direction. Avoid regions of conflict <Parallel tracking and mapping>
(e.g. close parallel edges). Deal with http://www.openslam.org/ <for SLAM
multiple matches through robust algorithms mainly from robotics
estimation. Video at community>
http://mi.eng.cam.ac.uk/~ee231/bmvcmovie.a http://www.robots.ox.ac.uk/~SSS06/
i. <SLAM literature and some software in
18RANSAC [Fischler and Bolles 1981]. Matlab>
RANSAC fit. Random Sampling ANd Consensus. 40Recommended intro reading: Yaakov
Least squares fit. Gross “outliers”. Bar-Shalom, X. Rong Li, Thiagalingam
Select random sample of points. Propose a Kirubarajan, Estimation with Applications
model (hypothesis) based on sample. Assess to Tracking and Navigation,
fitness of hypothesis to rest of data. Wiley-Interscience, 2001. Hugh
Repeat until max number of iterations or Durrant-Whyte and Tim Bailey, Simultaneous
fitness threshold reached. Keep best Localisation and Mapping (SLAM): Part I
hypothesis and potentially refine The Essential Algorithms. Robotics and
hypothesis with all inliers. Automation Magazine, June, 2006. Tim
19OK but… Having rich descriptors or Bailey and Hugh Durrant-Whyte,
even multiple kinds of features may still Simultaneous Localisation and Mapping
lead to wrong data associations (SLAM): Part II State of the Art. Robotics
(mismatches). If we pass to the SLAM and Automation Magazine, September, 2006.
system every measurement we think is good Andrew Davison, Ian Reid, Nicholas Molton
it can be catastrophic. Better to be able and Olivier Stasse MonoSLAM: Real-Time
to recover from failure than to think it Single Camera SLAM, IEEE Trans. PAMI 2007.
won’t fail! Andrew Calway, Andrew Davison and Walterio
20[Williams, Smith and Reid ICRA2007]. Mayol-Cuevas, Slides of Tutorial on Visual
Use 3 point algorithm -> up to 4 SLAM, BMVC 2007 avaliable at:
possible poses. Verify using Matas’ Td,d http://www.cs.bris.ac.uk/Research/Vision/R
test. Camera relocalization using small 2D altime/bmvctutorial/.
patches + RANSAC to compute pose. Adds a 41Fin. Some Challenges. Deal with larger
“supervisor” between visual measurements maps. Obtain maps that are task-meaningful
and SLAM system. (manipulation, AR, metrology). Use
21[Williams, Smith and Reid ICRA2007]. different feature kinds on an informed
In brief, while within real-time limit do: way. Benefit from other approaches such as
Carry on. Also see recent work [Williams, SFM but keep efficiency. Incorporate
Klein and Reid ICCV2007] using randomised semantics and beyond-geometric scene
trees rather than simple 2D patches. Video understanding.
at
Introduction: Robot Vision Philippe Martinet Unifying Vision and Control Selim Benhimane Efficient Keypoint Recognition Vincent Lepetit Multi-camera and Model-based Robot Vision Andrew Comport Visual SLAM for Spatially Aware Robots Walterio Mayol-Cuevas.ppt
http://900igr.net/kartinka/informatika/introduction-robot-vision-philippe-martinet-unifying-vision-and-control-selim-benhimane-efficient-keypoint-recognition-vincent-lepetit-multi-camera-and-model-based-robot-vision-andrew-comport-visual-slam-for-spatially-aware-robots-walterio-252500.html
cсылка на страницу

Introduction: Robot Vision Philippe Martinet Unifying Vision and Control Selim Benhimane Efficient Keypoint Recognition Vincent Lepetit Multi-camera and Model-based Robot Vision Andrew Comport Visual SLAM for Spatially Aware Robots Walterio Mayol-Cuevas

другие презентации на тему «Introduction: Robot Vision Philippe Martinet Unifying Vision and Control Selim Benhimane Efficient Keypoint Recognition Vincent Lepetit Multi-camera and Model-based Robot Vision Andrew Comport Visual SLAM for Spatially Aware Robots Walterio Mayol-Cuevas»

«Переменные Visual Basic» - Присваивание переменным значений. Пример программного кода Visual Basic. Объявление переменных. Byte, short, integer, long, single, double – типы числовых значений. A = 216 b = -31576 c = 3.1415926 D = “visual basic” А = А - 10. Имена переменных. Переменные: тип, имя, значение. Типы переменных. Переменная.

«English for you» - Викторины Задания Игры Игровые упражнения Ты можешь выбрать уровень сложности. Что ты научишься делать и узнаешь. EuroTalk. ENGLISH FOR YOU. You are welcome! Слова и выражения по темам. Ты сможешь совершенствовать своё произношение. Узнать насколько хорошо ты усвоил материал тебе помогут: Все слова и выражения озвучены носителями языка.

«Подпрограммы в Паскале» - Подпрограммы в языке Паскаль. Текст процедуры. Определение подпрограммы. Типы данных. Возвращаемые значения. Ситуация. Понятие подпрограммы. Деление двух дробей. Глобальная и локальная переменные. Описания двух различных подпрограмм. Объявление функции. Список параметров. Способы подстановки аргументов.

«Sport ua» - Брендирование. Баннерная реклама. Пол. Активная аудитория. Активная, лояльная и качественная аудитория. Крупнейший спортивный сайт. Видео-чаты и видео-конференции. Возраст. В среднем в день пользователь проводит на сайте 15 минут и просматривает более 10 страниц. Социальная сеть футбольных болельщиков.

«Вычислительные системы сети и телекоммуникации» - Недостаток: Повышение быстродействия центральных устройств не дает общего повышения быстродействия. ENIAC содержал 18000 электронных ламп. Сентябрь 1981 года. По функциональным возможностям и размерам: Джордж Буль. Зарождение компьютерных сетей 1961 год. Новый компьютер Стив Возняк и Стив Джобс назвали Apple-I.

«MS PowerPoint» - Использование MS PowerPoint. Не трать свое время, воспользуйся готовыми шаблонами презентаций. Пример слайда с видеоклипом. Способы представления готовой презентации. Пример слайда с автофигурами. Объекты, размещаемые на слайде. Средство подготовки презентаций. Пример слайда с диаграммой. Определение.

Без темы

778 презентаций
Урок

Информатика

130 тем
Картинки
900igr.net > Презентации по информатике > Без темы > Introduction: Robot Vision Philippe Martinet Unifying Vision and Control Selim Benhimane Efficient Keypoint Recognition Vincent Lepetit Multi-camera and Model-based Robot Vision Andrew Comport Visual SLAM for Spatially Aware Robots Walterio Mayol-Cuevas