Aucun titre de diapositive

1
Nicolas Loménie
University Paris Descartes Ass. Professor
lomn_at_sip-crip.org
Current Research interests / projects

• Unorganized Point Set Analysis

• Geographical Object Detection on Satellite
  Images
• Poultry food analysis by visual aspects
• Augmented Reality

Collaborations
Ankara
2
Unorganized 3D Point Set Analysis
3
Global structuring

• Clustering algorithm

Local Refinement

• 2D Mesh Morphological operators (to be published
  in Pattern Recognition Letter journal 2008)?

3D Point sets
Facts
Expert system paradigm
an application-driven strategy
Knowledge base
Clustering algorithm EFKM
Heuristic Rules Morphological operators ...
Inference motor
4
Global structuring

• Clustering algorithm

Minimisation of
where d dissimilarity measure
5
Global structuration

• Clustering algorithm

Unsupervised Partition by Improved Fuzzy K-Means
with exponential distance
6
Global structuration

• Clustering algorithm

Good performances in case of - variability in
density - variability in shapes - variability in
size -gt perfect for stereoscopic data
7
How to determine the good number of objects
heuristic rules to refine strategy
The ADP criterion Morphological analysis of
point set clusters
8
Morphological filtering relying on Delaunay
triangulations
Nina Amenta's team
9
?-objects
Edelsbrunner's team
10
? optimal 3 x median maillage ?
?-opening
11
References 1. Main ideas in A generic
methodology for partitioning unorganized 3D point
clouds for robotic vision , Lomenie Computer and
Robot Vision (CRV'01 ex VI), London, Canada, pp.
172-181, May 2004 2. More formal in Morphological
Mesh Filtering and ?-objects, Lomenie, to be
published in PRL 2008
Morphological operators
( as sub-graphs of Del)?
Opening
Erosion
?-Complex
?-Complex
?-Erosion of order 1
?-opening of order 1
?-Erosion of order 2
Cluster analysis by successive erosions
12
Useful for segmentation
13
(No Transcript)
14
(No Transcript)
15
Useful for reconstruction
16
Control of the visual aspects of poultry food
according to the poultry food behaviour
References R.Hachemi, N.Vincent, N.Loménie
Control of the visual and tactile aspects of
poultry food according to the poultry food
behavior by image analysis. 8th International
Conference on Quality Control by Artifical Vision
(QCAV2007), Le Creusot, France, (May 2007).