Anytime learning for the NoSLLiP tracker

We propose an anytime learning procedure for the Sequence of Learned Linear Predictors (SLLiP) tracker. Since learning might be time-consuming for large problems, we present an anytime learning algorithm which, after a very short initialization period, provides a solution with defined precision. As SLLiP tracking requires only a fraction of the processing power of an ordinary PC, the learning can continue in a parallel background thread continuously delivering improved, i.e. faster, SLLiPs with lower computational complexity and the same precision. The proposed approach is verified on publicly-available sequences with approximately 12,000 ground-truthed frames. The learning time is shown to be 20 times smaller than standard SLLiP learning based on linear programming, yet its robustness and accuracy is similar. Superiority in the frame-rate and robustness in comparison with the SIFT detector, Lucas-Kanade tracker and Jurie's tracker is also demonstrated.