MURI page1

Landmark Aquisition
and Environment Identification
for Visual Urban Navigation



Yutaka Takeuchi and Martial Hebert

The Robotics Institute
Carnegie Mellon University

http://www.cs.cmu.edu/~takeuchi/landmarks.html


page2

Approach

Visual Model Acquisition : Models are collections of images with similar feature distributions.


page3

Motivation


page4

Feature Distributions

Feature Distributions for 16 subimages


page5

Feature Distributions : Typical Examples


Original image





Color image after histogram quantization and normalization





Edge Image





page6

Grouping


page7

Grouping Example


Distance matrix for a 145-images training sequence;
darker points correspond to lower distances, the right images shows the distance matrix for the first 50 images.


page8

Image Transformations



page9

Image Registration


original image and registered image


page10

Matching



page11

On-Vehicle Experimentation

  • Tested on over ten sequences from vehicle
  • Algorithms exercised under difficult test conditions:
    • Non-modeled sequences
    • Large variation in illumination
    • Large variations in viewpoint
  • Tuned to minimize mis-classification
  • Sample statistics:
    • Matching performance in 12 sequences:
  •  

    Recognized

    Rejected

    Mis-Classified

    Model Images

    77.4%

    22.6%

    0%

    Non-Model Images

     

    99.7%

    0.3%


    page12


    page13

    Variation in Images

    Images from model group:

    Images recognized under large variation in viewpoint and illumination:


    page14

    Future Developments

  • Integration to vehicle system and demonstration
  • Optimization of time performance for practical operation
  • Exploration of use of Omnicamera