Skip to main content

Dr. Lin Yang; Statistics Colloquium Series

Date:
-
Location:
MDS 220

 

Refreshments: MDS 312, 3:30-4:00 pm
 
Title:
Collaborative Tracking and Robust Dictionary  Learning Using Sparse Representation
 
Abstract:
In this talk, I will first present a robust, fast and accurate 3D 
tracking algorithm: Prediction Based Collaborative Trackers (PCT). In 
PCT, a novel one-step forward prediction is introduced to generate the 
motion prior using motion manifold learning. Marginal space learning 
based boosting is used to speed up both the training and testing 
procedures. PCT is completely automatic and computationally efficient.
It requires less than 1.5 seconds to process a 3D volume which contains 
millions of voxels.
 
Although PCT can be used to track the 2D/3D object accurately and 
rapidly, it is not online updated. We have developed a robust online 
learning based tracking algorithm using a local sparse appearance model 
(SPT). A novel sparse representation-based voting map and sparse 
constraint regularized mean-shift are fused for adaptive object tracking 
through online learning.
 
In this talk, I will also introduce a novel dictionary learning algorithm 
with a locally constrained sparse representation, called K-Selection. I 
will also cover part of our ongoing research in developing a novel and 
robust K-selection (RKS) dictionary learning algorithm which is not only 
sparse but robust to outliers. These general algorithms are demonstrated 
for tracking, background substractiona and recognition, and can be used 
for statistical analysis as well