Real-Time High Resolution Photogrammetry

John Morris, Georgy Gimel’farb and

Patrice DelmasCITR, Tamaki Campus,University of Auckland

Applications

Navigation Collision avoidance in traffic Manoeuvring in dynamic environments

Biometrics Face recognition Tracking individuals

Films Markerless motion tracking

Security Intelligent threat detection

Civil Engineering Materials Science Archaeology

Why Stereo?

Passive Suitable for dense environments

Wide area 3D data can be acquired at high rates

Textureless regions cause problems Active illumination can resolve these

Photogrammetry Laboratory - Resources

Commercial structured lighting system Slow, limited depth of field Very accurate (~0.3mm) within its range (1-5m) Provides ‘ground truth’ for other techniques

Advanced structured lighting techniques Efficient labelling patterns Hybrid systems

• Structured lighting/Active illumation/Stereo High resolution cameras

4 x Canon digital SLR (2 x 6Mpixel, 2 x 8Mpixel) 2 x Baesler 25fps, 2 x Pixelink 25fps 4 x Allied Vision (2 with IR capability) Precise alignment, stable baselines

High speed cameras 2 x Fastec 250 fps

FPGA development kits Altera Development Kits Under development: Firewire interface – 2 x medium resolution

video cameras

Photogrammetry Lab

Stereo Canon digital

SLRs – 6 Mpixels

Low distortion lenses

Stable optical base

Precise alignment

Provision for verging optics

Target: Accurate 3D environment maps

Projects

Real time stereo vision Implementation of Symmetric Dynamic Programming Stereo

on FPGA hardware Real time precise stereo vision

• Faster, smaller hardware circuit• Real time 3D maps• 1% depth accuracy with 2 scan line latency

Stereo Applications Collision avoidance for automobiles and robots Face recognition via 3D models

• Fast model acquisition via stereo pairs

Resin flow in composite laminate formation Motion capture – athletes, actors

Projects

High Quality Rendering Scene rendering (movie quality) acceleration with

FPGA hardware Next generation animated movie requirements:

• Rendering farms with 1000’s of processors• Power supply alone becomes a major problem!

Custom hardware attached processors• Computationally intensive task• Speed ups of 10 or more• Same work with fewer processors

… and fewer power cuts!

Example Project: Composite Formation

Resin drawn by vacuum into flat bag containing fibre mat Problem: Measure shape of advancing resin wavefront Solution:

Stereo with verging camera axis configuration Process time series of images

• Transform and measure phase shift

Composite Materials Laboratory

Tamaki Campus Same building as Wine Science

8Mpixel Canon digital SLRs on stable base with precise alignment

Resin wavefront Profiles along the mould at 1, 2, 3, 4 minutes Note the resolution! Left scale is mm.

-1.2

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

mm

minute 1

minute 2

minute 3

minute 4