Re-identificando alvos de rastreamento

Abstract

This work tackle the task of target re-detection in the context of visual tracking. We propose the integration of correlation filter trackers along with Support Vector Machine classifiers to build a long-term appearance model capable of detecting tracking failures and recover from them. We perform extensive testing on the OTB dataset and analyse the results both quantitatively and qualitatively, providing evidence that the proposed method results on a more robust tracker.