ECOVISION Early Cognitive Vision

1
ECOVISION - Early Cognitive Vision

• The Theoretical Basis of ECOVISION
• Recursive Mid-Level Vision
• Application of Gestalt Laws in Computer Vision
• Optic Flow Gestalts
• Stereo Gestalts
• Results
• Fast Pre-Processing
• Gestalts in Flow and Stereo

2
Hierarchical-Recursive Image Processing
Pixels
Data and Noise reduction Extraction of
Meaningful Information (first steps)
Low-Level Vision
Features
Spatio-temporal Context Grouping,
Segmentation, Task-dependent Attention Self-Emerge
nce of Entities
Primitives
Mid-Level Vision
Gestalts
High-Level Vision (Cognition)
Higher Cognitive Aspects Reasoning
Objects
3
Examples
Features Orientation and Flow dense but noisy
4
Examples
Gestalt Meaningful entity that emerges from
cooperative, recursive processing.
This is not a traffic sign (according to R.
Magritte)
5
Properties of the Primitive Image Representation

• Local
• Condensed
• Sparse
• Rich
• Adaptive

6
Some more use of Primitives Symbolic
Pointillism
Glasgow,
Lighthouse Tower Exhibition in the context of the
ECVP
A cortical hyper- column representation
7
Examples of a Motion Gestalt
8
FPGA-based Motion Analysis Front-End

• Overtaking Sequence

9
Prototyping platform
10
Emergence of 3D Gestalts
3-D Gestalts emerge at high-energy locations
from local matches together with global
RBM properties
11
Summary

• Advanced Early Cognitive Image Processing for
  Mid-Level Vision
• Fast Front-End Processing
• Recursive, self-emergence of visual Gestalts
• Consortium
• F. Woergoetter Stirling
• Marc vanHuelle, Leuven
• Eduardo Ros, Granada
• Alan Johnston, London
• Markus Lappe, Muenster
• Silvio Sabatini, Genova
• Martin Muehlenberg, HELLA Hueck KG, Lippstadt