Automatic DTM Generation from ThreeLineScanner TLS Images

1
Automatic DTM Generation from Three-Line-Scanner
(TLS) Images
Armin Gruen, Zhang Li Institute of Geodesy and
Photogrammetry Federal Institute of Technology
(ETH) Zürich agruen, zhangl_at_geod.baug.ethz.ch,
www.photogrammetry.ethz.ch
1. Introduction 2. The TLS system 3. Matching
considerations 4. Sensor model for
triangulation and matching 5. Matching
ApproachRelaxation, MPGC, GCMM 6.
Experimental results 7. Conclusions
2
Introduction

• Cooperation IGP, ETH Zürich Starlabo
  Corporation, Tokyo
• Digital aerial cameras Need for
  automated procedures
• Image matching Possibilities not exploited yet
• TLS (Three-Line-Scanner) / Matching concepts
• 3-fold image coverage
• approx. orthogonal in flight direction
• Multi-image with geometrical constraints
  (MPGC)
• Multi-patch with neighbourhood and
  geometrical constraints (Relaxation
  GCMM)
• Goal Generate high quality DSM from TLS imagery

3
TLS System
4
(No Transcript)
5
Matching considerations
Results from commercial systems 1.5 to 16
times worse than manual measurements
Large number of big (and small) blunders
-----show statistics--- Results
imagescale-dependent Postediting Cannot solve
all problems
6
Problems in DTM generation
(a) Little or no texture (b) Distinct object
discontinuities (c) Local object patch is no
planar face in sufficient approximation (d)
Repetitive objects, incl. vegetation (e)
Occlusions (f) Moving objects, incl. shadows (g)
Multi-layered and transparent objects (h)
Radiometric artifacts, like specular reflections
and others (i) Reduction from DSM to DTM
Area-based, feature-based, relational matching
(advantages/disadvantages) Our matching strategy
Combination of algorithms
(Addresses (a) - (f))
7
(No Transcript)
8
Sensor model for triangulation matching
9
Collinearity conditions

• Triangulation 3 trajectory models
• Direct georeferencing with stochastic exterior
  orientations
• Piecewise Polynomials with kinematic model up to
  2nd order and stochastic 1st and 2nd order
  constraints
• Lagrange Polynomials with variable orientation
  fixes

10
Matching Approach

• Combination of
• Relaxation Matching ( approximations)
• MPGC.. Multi-photo Geometrically Constrained
  Matching
• GCMM.. Geometrically Constraint Multi-patch
  Matching

11
GC candidate search
12
GC Multi-patch Matching (Gruen, 1985b, Rosenholm,
1986,
Rauhala, 1988, Li, 1989)
TLS geometrical constraints
weighted
Image subdivision 11 x 11 grid meshes,
overlapping
13
Smoothness Constraints
With weights w l / Tex(i)
14
MPGC (Gruen 1985a) Feature point extraction by
interest operator
with constraints
without constraints
15
Experimental results
Results of triangulation
16
Matching Version
G_Rex Grid point matching based on
relaxation G-RexGCMM Refined grid point
matching of relaxation by GCMM G_RexF_MPGC
Combination of relaxation matching refined by
modified MPGC and feature point matching based on
MPGC
17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
Min. difference -0.44 m Max. difference 0.40
m Mean difference 0.01 m RMS error (z) 0.15
m
21
Conclusions

• Development of novel methods and software for TLS
  (Starlabo Corp., Tokyo)
• TLS sensor model 3 options ( experiments)
• Image matching New matching strategy
  (Relaxation, MPGC, GCMM) Bridge areas with
  low texture utilize image features
• RMSE (z) 0.15 m, (5.6 cm pixelsize)
• Main problem Small blunders
• In development Semi-automated matcher for 3D-
  city modeling