Tracking
Tracking
Problem
Track an object in a video sequence.
use a synthetic model of the object and a synthetic viewing model to represent the real scene.
Update model and view parameters so that the synthetic image matches the real image as close as possible.
See below for notes on posing this in terms of
tracking visualization features.
Givens
- video of object
- knowledge about the object
- possibly some subset of viewing and/or model parameters
- possibly the initial values of the unknown parameters
Preprocessing
- Build a synthetic model of the object
- Build a synthetic camera and projective viewing model
- If initial parameter values are unknown, estimate them. This is a global optimization problem and could be done by random sampling, simulated annealing, etc.
- Determine the method used to estimate and evaluate parameter adjustments.
- Determine a fitness function to be used to match the synthetic image with the real image.
Task
Use local optimization techniques to estimate adjustments to unknown parameters so as to reduce the difference between the projected real object and the projected synthetic model
Possible Parameters
- translation and rotation of rigid object
- camera position and orientation
- internal camera parameters such as focal length
- object shape parameters (fixed, but unknown)
- object shape parameters (time varying)
Relevant Literature (to be expanded)
Ultimate Objective
- Use a synthetic articulated figure fit to an image of a person in the video
Visualization Form
(from Roger)
Extend this to tracking features from visualization algorithms.
Givens:
- 2D visualizations of slices through the data.
- Knowledge about the visualization algorithm
Preprocessing
- Determine the set of visualization slices
- Determine the visualization parameters