TorrBreg-eccv02 - Space-Time Tracking Lorenzo Torresani and...

Info iconThis preview shows pages 1–3. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Space-Time Tracking Lorenzo Torresani and Christoph Bregler Computer Science Department, Stanford University, Stanford, CA 94305, USA { ltorresa, bregler } @cs.stanford.edu Abstract. We propose a new tracking technique that is able to cap- ture non-rigid motion by exploiting a space-time rank constraint. Most tracking methods use a prior model in order to deal with challenging local features. The model usually has to be trained on carefully hand- labeled example data before the tracking algorithm can be used. Our new model-free tracking technique can overcome such limitations. This can be achieved in redefining the problem. Instead of first training a model and then tracking the model parameters, we are able to derive trajectory constraints first, and then estimate the model. This reduces the search space significantly and allows for a better feature disambiguation that would not be possible with traditional trackers. We demonstrate that sampling in the trajectory space, instead of in the space of shape con- figurations, allows us to track challenging footage without use of prior models. 1 Introduction Most of the tracking techniques that are able to capture non-rigid motion use a prior model. For instance, some human face-trackers use a pre-trained PCA model or parameterized 3D model, and fit the model to 2D image features. Com- bining these models with advanced sampling techniques (like particle filters or multiple hypothesis approaches) result in algorithms capable of overcoming many local ambiguities. There are many cases where a prior-model is not available. In fact, often the main reason for performing tracking is to estimate data that can be used to build a model. In this case, model-free feature trackers have to be used. Unfortunately many non-rigid domains, such as human motion, contain challenging features that make tracking without a model virtually impossible. Examples of such features are points with degenerate or 1D texture (points along lips and eye contours, cloth and shoe textures). We propose an innovative model-free tracking solution that can overcome such limitations. This can be achieved in redefining the tracking problem. 1. Traditional Tracking : Given M → Estimate α : Standard model-based approaches assume a known (pre-trained) parameter- ized model M ( α ). The model M stays constant over the entire time sequence (for example M might coincide with a set of basis-shapes). The parameters α change from time frame to time frame (for example the interpolation coef- ficients between the basis shapes). Traditional tracking solves by estimating frame by frame the parameters α (1) ..α ( F ) that would fit M ( α ) to the data. 2. Reverse-Order Tracking : Estimate α → Estimate M : Our new technique first estimates the α (1) ..α ( F ) without knowing the model....
View Full Document

This note was uploaded on 06/13/2011 for the course CAP 6412 taught by Professor Staff during the Spring '08 term at University of Central Florida.

Page1 / 12

TorrBreg-eccv02 - Space-Time Tracking Lorenzo Torresani and...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online