This webpage is devoted to my Phd project `The Electronic Stockmans Eye'
and details work carried out between 1997 and 2000.
This project involved the modeling of livestock shape and behaviour using machine
vision techniques. Much of my initial work involved variations on the
Point Distribution Model (PDM) developed by Cootes and Taylor at Manchester
University. Later work was based on variations on the Condensation algorithm
of Isard and Blake from Oxford University. This work included :
Developing method of automatically determining landmark point positions
and numbers using a straight line approximation method.
Developing a PDM based around a vector representation.
Developing methods for separating out inter and intra-class variation in
PCA based models.
Developing new object tracking methods.
Developing a method of modeling and classifying temporal behaviour.
Object Modeling
Livestock contours are modeled by multiple models based on a piecewise
linear representation. The variation within each model is separated into
front legs, rear legs and inter-animal components in a hierarchical manner.
Cow Tracking in Action
Models are fitted to unseen sequences using an intelligent search algorithm
based on a temporal model and a particle filtering scheme similar to the
CONDENSATION algorithm.
Demos
The `Virtual Cow' is created by feeding the output of the temporal predictors
back into the inputs and stochastically selecting the next states.
Some tracking results (the maximum likelyhood result is shown).
Related Publications
Magee D. and Boyle R.
Building Shape Models from Image Sequences Using
Piecewise Linear Approximation.
Proc. British Machine Vision Conference,
pages 398-408. BMVA, September 1998. Available as a Research
Report.
Magee D. and Boyle R.
Feature tracking in real world scenes (or how
to track a cow).
Proc. IEE Colloquium on Motion Analysis and Tracking,
IEE, May 1999. Available as a Research Report.
Magee D. and Boyle R.
Building Class Sensitive Models for Tracking Applications.
Proc.
British Machine Vision Conference, BMVA, September 1999. Available
as a Research Report.
Magee D. and Boyle R.
Spatio-temporal Modeling in the Farmyard Domain.
Proc.
IAPR International Workshop on Articulated Motion and Deformable Objecs
(AMDO'2000), Palma de Mallorca, Spain, September 2000, pp83-95.
Available as a Research
Report.
Magee D. and Boyle R.
Detecting lameness in livestock using resampling
Condensation and multi-stream cyclic hidden markov models.
Proc.
British Machine Vision Conference
, BMVA, September 2000, pp332-341.
Available as a Research
Report.Winner of Best Industrial
Paper Award
Magee D. and Boyle R.
Detecting lameness in livestock using resampling
Condensation and multi-stream cyclic hidden markov models.
Image and Vision Computing
, vol 20(8), pp 581-594, 2002.
Available as a Research Report.
