Robust coding in a global subspace model and its collaboration with a local model for visual tracking

摘要: The sparse representation-based trackers has attracted much attention in the research community due to its superior performance spite of computational complexity. But assumption that coding residual follows either Gaussian or Laplacian distribution may not accurately describe practical visual tracking scenarios. To deal with such issues as well improve tracking, a novel generative tracker is proposed Bayesian inference framework by introducing robust (RC) into PCA reconstruction. Also, it collaborate global and local subspace appearance models enhance performance. Further, RC distance differentiate candidate samples from subspace, observation likelihood defined based on both distances. In addition, occlusion map generation model update mechanism are proposed. quantitative qualitative evaluations OTB-50 VOT2016 dataset demonstrate method performs favorably against several methods particle filter framework.