Robust Visual Tracking with Discrimination Dictionary Learning

Algorithm 1

Proposed tracking algorithm.

In the first frame, manually select the tracked target state ; collect training samples

; learn a dictionary according to the learning scheme

in Section 2.2; sample particles with equal weights as .

Input: t-th video frame.

Resample particles according to motion model .

Extract feature vectors according to .

for to do

Compute observation likelihoods via Eqn. (15).

Update particle weight via Eqn. (3).

end

Compute target state with Eqn. (2).

Extract feature vector according to .

Update the training samples with .

Learn dictionary according to Eqns. (11) and (13) in Section 2.2.

Obtain dictionary .

Return .