Viewpoint Invariant Pedestrian Recognition with an Ensemble of Localized Features

摘要: Viewpoint invariant pedestrian recognition is an important yet under-addressed problem in computer vision. This likely due to the difficulty matching two objects with unknown viewpoint and pose. paper presents a method of performing using efficiently intelligently designed object representation, ensemble localized features (ELF). Instead designing specific feature by hand solve problem, we define space our intuition about let machine learning algorithm find best representation. We show how both class representation discriminative model can be learned AdaBoost algorithm. approach allows many different kinds simple combined into single similarity function. The evaluated dataset results are shown superior all previous benchmarks for reacquisition pedestrians.