High performance person re-identification via a boosting ranking ensemble

Abstract

Person re-identification is a very challenging task in computer vision due to severe appearance changes of a person across non-overlapping camera views. It is thus inadequate in most realistic re-identification scenarios to assess similarity using a single metric in a single feature space. Ensemble methods have been proven effective on improving the recognition rate. However, robust combination remains challenging due to the incompatibilities between different distance metrics. In this paper, we propose a novel framework for person re-identification called Boosting Ranking Ensemble (BRE), which adaptively assembles features and metrics from a ranking perspective. To further tackle the overfitting problem, we explore an effective rectifier loss in the proposed BRE framework to alleviate the influence of the misclassified samples. Extensive experimental analyses and evaluations on three commonly adopted benchmarks demonstrate the effectiveness of the proposed method, with superior performance over many state-of-the-art methods.