Farfield Terrain Classification

1
Far-field Terrain Classification

• Life in the Atacama Design ReviewDecember 19,
  2003
• Alec Avedisyan
• CMU EPFL

2
Description and Motivation

• Make a classification system, based on the
  rovers sensors, to identify obstacles and target
  which are in the middle range.
• Traversability of the terrain
• Physical properties

3
General Idea Illustration
4
Key Requirements
Classification system has to fill the gap between
two information's ? - near field (0-7m) ? -
satellite imagery (30-infinite) ?
5
Key Requirements

• Two approaches
• Semantic/Topologic
• A manner of describing the space by unspecified
  areas (or object)
• Image processing
• Image sequences, Features detector, Image filter,
  etc
• Geometric
• Modeling geometrical space, volumes, distances
• Stereo cameras
• 1D or 2D Laser
• Other sensors

6
General Idea Semantic Illustration
7
General Idea Semantic Illustration
Infinite Distant Far Near
8
Toposementic - Implementation

• Features detector
• Filter (high frequencies, )
• Edges detector
• Color-based detector
• Fast Fourier Transform

9
Geometric - Implementation
Stereo ? Yes, but Laser ? Yes, but
10
Implementation Issues

• Determine algorithms for
• Stereo features detection for wide baseline
• Distance extraction
• Texture indexation
• Areas categorization
• Traversability estimator