Download presentation
Presentation is loading. Please wait.
Published byColeen Lang Modified over 9 years ago
1
Online Tracking of Outdoor Lighting Variations for Augmented Reality with Moving Cameras Yanli Liu 1,2 and Xavier Granier 2,3,4 1: College of Computer Science, Sichuan University, P.R.China 2: INRIA Bordeaux Sud-Ouest, France 3:LP2N (CNRS, Univ. Bordeaux, Institut d'Optique) 4:LaBRI (CNRS, University of Bordeaux)
2
20 12 Augmented reality mobility Motivation
3
20 12 TTwo consistency Geometric consistency DDevices CCamera position GPS, UWB, Omnisense WiFi, cell information CCamera pose Linear accelerometers TTracking via computer vision [[Cornelis et al. LNCS 2001, zhang et al. CVPR 2007, Xu et al. image and Vision Computing 2008] Illumination consistency ooutdoor lighting is largely dependent on weather and time
4
20 12 Two problems Online process first step toward real-time solutions Moving viewpoints Handhold camera jitter
5
20 12 Previous Work Markers or lighting probes [Debevec Siggraph’ 98, Agusanto ISMAR’03, Kanbara ICPR’04, Hensley I3D’07] too dense sampling our method does not require any supplemental devices Debevec Siggraph’ 98
6
20 12 Previous Work Three components of shading BRDF geometry lighting Fix other one or two components [Wang PG’02, Li ICCV’03, Hara PAMI’05, Andersen ICPR’06, Sun ICCV’09] 3D reconstruction controlled environment (indoor or lab) [Wang PG’02] original image rendered image
7
20 12 Previous Work Time-lapse outdoor video analysis [Sunkavalli Siggraph’07, Sunkavalli CVPR 08] take whole video sequence as input Post-processing [Sunkavalli Siggraph’07]
8
20 12 Previous Work Learning based outdoor illumination estimation [Liu TVC’09, Liu CAVW’10, Xing C&G’11] offline stage learning fixed viewpoint Liu CAVW’10 moving viewpoints
9
20 12 Our Method Key ideas Tracking illumination variation by tracking feature points Feature points tracking is error prone. Select reliable feature points using global illumination constraint and spatial-temporal coherency.
10
20 12 Outdoor lighting [Sunkavalli SIG’07, Sunkavalli CVPR 08, Madsen InTech 2010] the sunlight directional light colored intensity sun direction the skylight ambient light colored intensity Illumination and BRDF model
11
20 12 Neutral reflection model [Lee PAMI’90, Montoliu LNCS’05, Eibenberger ICIP 2010, ICCV 2011] the color of the specular reflection is the same as the color of the incident lighting. Phong-like model
12
20 12 System Initialization Tracking illumination variation by tracking feature points 3D geometry vs normals planar feature points KLT feature-points clustered feature-points first frame plane segmentation [Hoiem IJCV’07] mean-shift color segmentation threshold-based Shadow detection
13
20 12 System Initialization BRDF initialization pixels difference at in sun lit regions depend on specular parameters and : Assuming piecewise constant, and Spatially varying diffuse
14
20 12 Energy function Outdoor lighting is nearly constant during time intervals less than 1/5 second. control the smooth degree of skylight Alignment-based weight Tracking Lighting Variation with Reliable Feature Points
15
20 12 Tracking Reliable Features and Their Attributes Feature points labeling Three attributes: Normal (plane, homography matrix) BRDF parameters Shadow situation Spatial & temporal coherency
16
20 12 Tracking Reliable Features and Their Attributes Feature points labeling t -1 t paired point is labeled in current point is not in shadow compute lighting
17
20 12 Results and Discussion Quantitative results PC: Intel i7 2.67GHz and 6GB RAM MATLAB Video resolution 640 480 Average fps and average number of feature points estimated on 1,000 frames
18
20 12 Results and discussion Quantitative results Average percentage of different steps in total computational cost
19
20 12 Results and Discussion Visual results Building scene Wall scene
20
20 12 Conclusion Fully image-based pipeline online tracking of lighting variations of outdoor videos. Manages lighting changes and misalignment of feature points Ensure a stable estimation on a sparse set feature points.
21
20 12 Limitations and Future Work Rough shadow detection 3D reconstruction vs shadow detection Sun-lit features Initialization automatic initialization: easy but may fail in some cases manual initialization: may be tedious for a non-expert user. Semi-assisted initialization
22
20 12 Limitations and Future Work Tracking independently on R, G, and B channels priori model of outdoor illumination color or spectra difficult to optimization The first step of a long march to a seamless and real-time AR solution for videos with moving viewpoints.
23
20 12 Thanks for your attention! Questions?
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.