Dynamic Resource Allocation by Ranking SVM for Particle Filter Tracking

We propose a dynamic resource allocation algorithm based on Ranking Support Vector Machine (R-SVM) for particle filter tracking. We adjust the number of observations in each frame adaptively, where tracker performs measurement for a subset of particles to preserve mode locations in the posterior and allocates the rest of particles to maintain the diversity of the posterior without actual measurements. The number of measurements is determined by a ranking classifier, which evaluates the quality of each particle and counts the number of good ones. The ranking classifier is trained by R-SVM algorithm and universally applicable to every object and sequence because it is learned based on observation likelihoods, not image features. Our algorithm is useful to reduce observation cost and improve sampling quality in particle filtering. We integrated the proposed technique into l1 minimization tracking based on sparse representation and validated its effectiveness with several challenging videos.