Project Leader
Alexander Milward Arthur
MSc, School of Informatics, University of Edinburgh |
Project Supervisor
Sethu Vijayakumar,
PhD
IPAB, School of
Informatics, Univ. of Edinburgh |
Project Title
Object Tracking through Adaptive Cue Integration
Project Goal
The problem domain of object tracking is explored by way of a
discussion of existing frameworks. Democratic integration (Triesch
& von der Malsburg, 2001) is presented as a non-probabilistic
method. A preferrable method is one which learns the appropriate
parameters from raw data provided only by the camera, rather than
setting the parameters heuristically in advance as in Democratic
Integration. Thus, probabilistic frameworks are investigated, starting
first with the Transformed Hidden Markov Model (Jojic, Petrovic, Frey,
& Huang, 2000). This model does not include multiple cues, so we
turn to a probabilistic model which does integrate multiple cues albeit
for segmentation rather than tracking (Hayman & Eklundh, 2002).
Finally, we modify an existing framework, Bayesian Modality Fusion
(Toyama & Horvitz, 2000; Sherrah & Gong, 2001), such that cues
are permitted to adapt over time.
- Democratic Integration (Triesch & von der Malsburg, 2001)
- Modality Reports: Color, Motion, Position, etc.
- Voting System
- Testing and Results
- Transformed Hidden Markov Model (Jojic, Petrovic, Frey, & Huang, 2000)
- Model Implementation
- Testing and Results
- Probabilistic Approach to Cue Integration (Hayman & Eklundh, 2002)
- Model Implementation
- Testing and Results
- Bayesian Modality Fusion (Toyama & Horvitz, 2000) with Adaptive Cues
- Theoretics, Model Creation
- Model Implementation
- Testing and Results
Triesch, J., & von der Malsburg, C. Democratic integration: Self-organized integration of adaptive cues. Neural Computation, 13(9):2049-2074, 2001.
Jojic, N., Petrovic, N., Frey, B.J., & Huang T.S. Transformed
Hidden Markov Models: Estimating Mixture Models of Images and Inferring
Spatial Transformations in Video Sequences. Proc. IEEE Conf. Computer Vision and Pattern Recognition, 2000.
Hayman, E., & Eklundh, J.-O.. Probabilistic and voting approaches to cue integration for figure-ground segmentation. European Conference on Computer Vision, 2002.
Toyama, E., & Horvitz, E. Bayesian Modality Fusion: Probabilistic
Integration of Multiple Vision Algorithms for Head Tracking. Proceedings of Fourth Asian Conference on Computer Vision, 2000.
Sherrah, J., & Gong, S. Continuous global evidence-based Bayesian
Modality Fusion for simultaneous tracking of multiple objects. Porc. IEEE International Conference on Computer Vision , pages 42-49, Vancouver, Canada, 2001.
Project Timeline
|
Time Frame |
Task (completed or
scheduled) |
| 1
Jun. 04 - 15 Jun. 04 |
Modality Reports |
| 1
Jun. 04 - 15 Jun. 04 |
Voting System |
| 15
Jun. 04 - 18 Jun. 04 |
Testing and Results of Democratic Integration |
| 18
Jun. 04 - 2 Jul. 04 |
Implementation of Transformed Hidden Markov Models |
| 5
Jul. 04 - 9 Jul. 04 |
Testing and Results of Transformed Hidden Markov Models |
| 9
Jul. 04 - 23 Jul. 04 |
Theoretics and Model Creation of Bayesian Modality Fusion with Adaptive Cues |
| 23
Jul. 04 - 30 Jul. 04 |
Implementation of Bayesian Modality Fusion with Adaptive Cues |
| 30
Jul. 04 - 6 Aug. 04 |
Testing and Results of Bayesian Modality Fusion with Adaptive Cues |
| 6
Aug. 04 - 1 Sep. 04 |
Dissertation (note: will also be written concurrently with tasks above) |
Project Results & Conclusions
Thesis
|
