Individual Zebra Identification

jabgoldfishAI and Robotics

Oct 19, 2013 (3 years and 10 months ago)

102 views

Individual Zebra Identification

Hans Krijger


Supervised by
Shaun Bangay

and
Greg Foster

The Problem


Tracking individual zebra is useful


Feeding, breeding and migration trends


Social interaction


Wildlife management & ecological assessment



Traditional methods have drawbacks


Expense (Radio transmitters)


Time (Photographic methods)


Trauma during capture (Tranquillisation)


The Solution


Computer aided matching


Fast


Cost
-
effective


Non
-
intrusive



Draw on existing knowledge


Image processing methods


Fingerprint verification model


Machine vision pipeline


Solution Model


Two main stages


Important to standardise at each step


Reduce variation


Improve accuracy

Enhancement

De
-
interlacing by averaging

Binarisation

Skeletonisation



Iterative sequential thinning



Uses mathematical morphology



Direction and rotation invariant

Matching Techniques


Image Overlay


Simple: scaled images


Warp: new co
-
ordinate system


Vector Overlay


Neighbourhood scoring


Weighted Factor


Branch, curvature, intersection
score


Correlations of these


Statistical


Discriminant analysis


Principal component analysis


Implementation Specifics


Java


Cross
-
platform


Flexible and powerful


Only basic imaging
support



ImageJ


Open source


Improves on Java
imaging


Provides basic image
processing methods


Pattern Extraction and

Matching Interface

Demonstration


Dual implementations


Simple:

aimed at end user


Functional
:

used for experimentation

Results

Conclusions


Image preparation


Enhancement & binarisation
successful


Skeletonisation
can be improved


Segmentation
should be automated



Image Analysis


Contour tracing
successful


General feature extraction
successful


Matching needs
further investigation


Questions?