57
Interactive Point-based Modeling of Complex Objects from Images Pierre Poulin (a,b) Marc Stamminger (a,c) François Duranleau (b) Marie-Claude Frasson (a) George Drettakis (a) (a) REVES, INRIA Sophia Antipolis (b) DIRO, Université de Montréal (c) University of Erlangen

Interactive Point-based Modeling of Complex Objects from Images

  • Upload
    lise

  • View
    24

  • Download
    3

Embed Size (px)

DESCRIPTION

Interactive Point-based Modeling of Complex Objects from Images. Pierre Poulin ( a,b ) Marc Stamminger ( a,c ) François Duranleau ( b ) Marie-Claude Frasson ( a ) George Drettakis ( a ) ( a ) REVES, INRIA Sophia Antipolis ( b ) DIRO, Université de Montréal ( c ) University of Erlangen. - PowerPoint PPT Presentation

Citation preview

Page 1: Interactive Point-based Modeling of Complex Objects from Images

Interactive Point-based Modeling of Complex Objects from Images

Pierre Poulin (a,b)Marc Stamminger (a,c)François Duranleau (b)

Marie-Claude Frasson (a)George Drettakis (a)

(a) REVES, INRIA Sophia Antipolis(b) DIRO, Université de Montréal

(c) University of Erlangen

Page 2: Interactive Point-based Modeling of Complex Objects from Images

Modeling Complex Objects

Page 3: Interactive Point-based Modeling of Complex Objects from Images

Modeling Complex Objects

• High visual complexity

• Time consuming

• Algorithms for specialized objects– e.g., plants, mountains, etc.

• Adaptive rendering

• Many applications need such objects

Page 4: Interactive Point-based Modeling of Complex Objects from Images

Key Observations

• Extracting complex models from photos is a very powerful approach

• Point-based representation is very effective for complex models– Efficient display and storage

• User interaction is beneficial when extracting quality models– Specify where details are needed

– Resolve some ambiguities

Page 5: Interactive Point-based Modeling of Complex Objects from Images

Image-based Point Modeling

• Images are very flexible– Reality-based (photos)– Acquisition is easy

Page 6: Interactive Point-based Modeling of Complex Objects from Images

Image-based Point Modeling

• Points are very flexible– Fast rendering (hardware support)– Adaptive rendering for interactive display

Stamminger

Page 7: Interactive Point-based Modeling of Complex Objects from Images

Image-based Point Modeling

• Points are very flexible– Hierarchical organization and levels of detail

Q-splat

Page 8: Interactive Point-based Modeling of Complex Objects from Images

Image-based Point Modeling

• Points are very flexible– Visual quality– Many recent advances

Deussen

Page 9: Interactive Point-based Modeling of Complex Objects from Images

Automatic ReconstructionImages

ReconstructionProcess

Constraints

3D Model

Image

Page 10: Interactive Point-based Modeling of Complex Objects from Images

Interactive ReconstructionImages

ReconstructionProcess

Constraints

3D Model

Image

User

new imagesrequantizerecalibrate

Page 11: Interactive Point-based Modeling of Complex Objects from Images

Interactive ReconstructionImages

ReconstructionProcess

Constraints

3D Model

Image

User

color comparisonsplausibility threshold

new depth mapszone of interest

Page 12: Interactive Point-based Modeling of Complex Objects from Images

Interactive ReconstructionImages

ReconstructionProcess

Constraints

3D Model

Image

Userrevalidate the pointsrequest more pointsdecimate the points

jitter the pointssample with patterns

hole filling

Page 13: Interactive Point-based Modeling of Complex Objects from Images

Interactive ReconstructionImages

ReconstructionProcess

Constraints

3D Model

Image

User

undo changesremove pointsadd polygons

Page 14: Interactive Point-based Modeling of Complex Objects from Images

Interactive Reconstruction

• Interactive display– 6 M points/sec. on a PIII 1GHz with GeForce3

• Efficient reconstruction algorithm– Test more than 1K points/sec.

• Simple and intuitive controls– Direct interaction with the points

Page 15: Interactive Point-based Modeling of Complex Objects from Images

Computer Vision Contributions

• 3D scanners

• Structured light

• Stereo – N-views

• Shape-from-X

• Volumetric

Page 16: Interactive Point-based Modeling of Complex Objects from Images

Volumetric Reconstruction

• Voxel coloring and Space carving– If a voxel is impossible, carved out of object– Silhouettes, transparency, shading– Photo-consistency

SeitzKutulakos

Page 17: Interactive Point-based Modeling of Complex Objects from Images

Image-based Polygon Modeling

• Academic: Façade, Rekon, Reality

• Industry: RealViz, Canoma, Photomodeler

Façade

Page 18: Interactive Point-based Modeling of Complex Objects from Images

Image-based Polygon Modeling

• Small polygonal scene (30-100 polygons)

• Extracted textures and illumination

Boivin

Page 19: Interactive Point-based Modeling of Complex Objects from Images
Page 20: Interactive Point-based Modeling of Complex Objects from Images

Input Images (4/14)

Page 21: Interactive Point-based Modeling of Complex Objects from Images

Input Images

• Digital camera: Canon EOS-DS30

• 1080x720 and 2166x1440

• Fixed aperture and shutter speed

• Try not to change zoom

• OpenGL and ray traced test scenes

Page 22: Interactive Point-based Modeling of Complex Objects from Images

Camera Calibration

Page 23: Interactive Point-based Modeling of Complex Objects from Images

Camera Calibration

• ImageModeler from RealViz

• Fiduciary marks placed around the object

• Interactive system

• Intrinsic and extrinsic camera parameters

Page 24: Interactive Point-based Modeling of Complex Objects from Images

3D Zone of Interest

Page 25: Interactive Point-based Modeling of Complex Objects from Images

Initial Random Points

Page 26: Interactive Point-based Modeling of Complex Objects from Images

Initial Random Points

• Generated randomly within the envelope

• More specific patterns discussed later

• Projection of a point in each photo

• Gather colors

Page 27: Interactive Point-based Modeling of Complex Objects from Images

Color Comparison

• Euclidean distance– RGB, CIE xy, CIE Luv, CIE Lab

– Speed vs. accuracy

• Color quantized images– Precomputed (ppmquantall or more sophisticated)

– Quantization only on projected zone of interest

– 32 to 128 colors

– Reduce shading variations

– Efficient test for color equality

Page 28: Interactive Point-based Modeling of Complex Objects from Images

C: 25%B: 50%

Dominant Color

A: 100%

Plausibility

100%33%

with visibility

Page 29: Interactive Point-based Modeling of Complex Objects from Images

Random Points with Depth Maps

Page 30: Interactive Point-based Modeling of Complex Objects from Images

Depth Maps

• Computed from the current set of points

• Updated on user demand

• With depth maps, can raise the plausibility threshold

• Generate more points within the object

• Re-evaluation of previously generated points

Page 31: Interactive Point-based Modeling of Complex Objects from Images

Clean-up Points

Page 32: Interactive Point-based Modeling of Complex Objects from Images

Clean-up Points

• In general– Increase color threshold and re-evaluate

• With good depth maps– Project in each image– Reject if point visible and color too different

Page 33: Interactive Point-based Modeling of Complex Objects from Images

Generate More Points

Page 34: Interactive Point-based Modeling of Complex Objects from Images

Generate More Points

• Randomly

• Stratified sampling based on voxels

• Point decimation based on voxels

Page 35: Interactive Point-based Modeling of Complex Objects from Images

Guide the Points

Page 36: Interactive Point-based Modeling of Complex Objects from Images

Guide the Points

• Smaller 3D sphere of interest– Generate more points– Eliminate all points

• 3D flood fill for branching patterns

• Patterns for planar surfaces

• Patterns for boundary surfaces

Page 37: Interactive Point-based Modeling of Complex Objects from Images

Filling with no Leaves

Page 38: Interactive Point-based Modeling of Complex Objects from Images

Filling with Leaves

Page 39: Interactive Point-based Modeling of Complex Objects from Images

Jitter the Points

Page 40: Interactive Point-based Modeling of Complex Objects from Images

Reprojection

Page 41: Interactive Point-based Modeling of Complex Objects from Images

Stepping through it again

Page 42: Interactive Point-based Modeling of Complex Objects from Images

Results

Scene Images Resolution Colors Points

Fruit bowl 13 512x512 - -

Soldier 13 2160x1440 64 118K

Snack 8 1440x960 64 120K

Ficcus 13 2160x1440 64 150K

Page 43: Interactive Point-based Modeling of Complex Objects from Images

Synthetic Fruit Bowl

ray tracingcolor points reprojection

Page 44: Interactive Point-based Modeling of Complex Objects from Images

Toy Soldier

color pointscolor points reprojection

Page 45: Interactive Point-based Modeling of Complex Objects from Images

Snack

Page 46: Interactive Point-based Modeling of Complex Objects from Images

Snack

Page 47: Interactive Point-based Modeling of Complex Objects from Images

Ficcus

Page 48: Interactive Point-based Modeling of Complex Objects from Images

Conclusions

• Point-based reconstruction of complex objects from images

• Tight integration– 3D color point representation– User-driven and/or automatic reconstructions– Interactive display

• Flexible to integrate most advances in computer vision

Page 49: Interactive Point-based Modeling of Complex Objects from Images

Findings

• First steps are encouraging, but objects are still of limited realism

• Information in photos is inspiring, but also difficult to analyse correctly

• How many things in a pixel?

• How many pixels and colors for an object?

Page 50: Interactive Point-based Modeling of Complex Objects from Images

Future Work

• Video sequences

• High dynamic range photos

• Shadows and shading in color comparison

• Extraction of limited BRDFs

• 3D texture synthesis of materials

Page 51: Interactive Point-based Modeling of Complex Objects from Images

Questions

• Did you…

• Is it…

• Can you…

• When…

• What…

• Where…

Page 52: Interactive Point-based Modeling of Complex Objects from Images

User Interaction in Modeling

• Specify regions of interest, thresholds, validity

• Control the visual quality

• Iterative refining process

• Guide the solution

• Automatic or interactive process

• Interactive display (6 M points/sec. GeForce3)

Page 53: Interactive Point-based Modeling of Complex Objects from Images

Image-based Point Modeling

• Difficulties with points – Visibility

• Holes in surfaces, size of points

• Filtering the representation and the texture

• Not our goal to fix these difficulties for now

Page 54: Interactive Point-based Modeling of Complex Objects from Images

LOD in Graphics

• Environment maps

• Billboards

• Textured polygons

• Layer-depth images

• Light field / lumigraph

Page 55: Interactive Point-based Modeling of Complex Objects from Images

3D Scanners

• Very good results in general

• Size of the scanner wrt object

• Costs

• Fixed illumination

Page 56: Interactive Point-based Modeling of Complex Objects from Images

Stereo - N views

• Camera calibration• Epipolar constraints• Color matching• 3D position and color• Difficulties

– Holes and occlusions– Sharp edges, noise, shading

• Infinity of shapes in general• Targeted for object recognition and collision avoidance• Only recently goal of photo-realism

Page 57: Interactive Point-based Modeling of Complex Objects from Images

Shape-from-X

• Silhouettes

• Shadows

• Focus/defocus

• Motion

• Shading

• etc.