Data Acquisition for Digital Reconstruction of 3D Objects and Environments

1
Data Acquisition for Digital Reconstruction of 3D
Objects and Environments

• Low Kok Lim
• lowkl_at_comp.nus.edu.sg
• January 12, 2006

2
Range Scanners

• A range scanner is a device that measures
  distances between itself and some surface points
  on an object or in its environment
• Produces range images, and info at each pixel
  is a range value
• The process of acquiring a range image is called
  a range scan
• Hence, range scanners

3
Range Scanner Example

• The DeltaSphere-3000 Laser 3D Scene Digitizer
• Made by 3rdTech
• Uses time-of-flight range sensing
• Mid-range, suitable for scanning indoor
  environments
• Effective range 1 ft to 50 ft

4
Range Image
3D model created from the above range image
5
Intensity Image
6
Color Information

• Color information of the surface points are
  captured using ordinary digital color camera
• Color images are then manually registered and
  mapped to the range data or 3D model
• Each surface point on model is finally mapped
  with ONE color

7
Example Colored Model
8
Example Colored Model
9
View-Dependent Color

• Observation Color on real surface points are
  actually view-dependent
• Need to capture the surface reflectance from many
  different directions
• Therefore, need to take many photographs
• Problem Need to automate registration of color
  images to 3D models, where color images are taken
  with untracked cameras
• Manual registration is too tedious and error-prone

10
Project 1

• Automatic Registration of Color Images to 3D
  Models
• Goal Reconstruct 3D digital models of objects
  and indoor environments with view-dependent
  surface reflectance
• In addition, to achieve very good registration,
  we need to deal with internal calibration errors
  (distortions) in range scanners and color cameras

11
Multiple Scans
single scan

• One scan is not enough
• occlusions, scanners limitations, etc.
• Multiple scans from different views needed to
  reconstruct model

reconstructed from 8 range images
12
Patching Up Holes

• Problem Model still has holes at regions that
  are difficult to scan with large long-range
  scanner
• Solution Use small hand-held, video-rate,
  short-range scanner to scan the holes

13
Project 2

• Real-Time Hand-Held 3D Range Scanning of Indoor
  Environments
• Goal provide real-time visualization of the
  scanning result to user
• Challenge requires real-time registration of
  range images and real-time 3D reconstruction and
  fast rendering

14
View Planning

• Multiple scans from different views needed to
  reconstruct more complete model
• Scanner views must be chosen carefully, so that
• 3D model satisfies reconstruction quality
  requirements
• minimize number of scans
• Must consider acquisition constraints
• e.g. visibility, limited FOV DOF, minimum
  clearance distance, etc.
• Manual view planning is difficult
• Need automated view planning

15
Automated View Planning

• A priori knowledge of environments geometry not
  available
• First view manually selected, and each subsequent
  view computed
• based on information from previous scans

16
Project 3

• High-Dimensional View Planning for Range
  Acquisition
• Goal Practical automated 5D view planning (5D
  pose 3D position pan tilt)
• Acquire a minimal set of range images so that the
  3D model reconstructed can satisfy specified
  quality requirements, and also to ensure
  acquisition constraints are satisfied during
  range acquisition

17
More Info?

• These PPT slides, and detailed project
  descriptions can be found at http//www.comp.nus.e
  du.sg/lowkl/projects/
• Email me at lowkl_at_comp.nus.edu.sg
• Talk to me (office S15 06-02)