Boosted human re-identification using Riemannian manifolds

This paper presents an appearance-based model to address the human re-identification problem. Human re-identification is an important and still unsolved task in computer vision. In many systems there is a requirement to identify individuals or determine whether a given individual has already appeared over a network of cameras. The human appearance obtained in one camera is usually different from the ones obtained in another camera. In order to re-identify people a human signature should handle difference in illumination, pose and camera parameters. The paper focuses on a new appearance model based on Mean Riemannian Covariance (MRC) patches extracted from tracks of a particular individual. A new similarity measure using Riemannian manifold theory is also proposed to distinguish sets of patches belonging to a specific individual. We investigate the significance of MRC patches based on their reliability extracted during tracking and their discriminative power obtained by a boosting scheme. Our method is evaluated and compared with the state of the art using benchmark video sequences from the ETHZ and the i-LIDS datasets. Re-identification performance is presented using a cumulative matching characteristic (CMC) curve. We demonstrate that the proposed approach outperforms state of the art methods. Finally, the results of our approach are shown on two further and more pertinent datasets.

► We propose a human appearance model for re-identification.
► Essential information is kept by Mean Riemannian Covariance (MRC) patches.
► Patch selection is based on a boosting scheme.
► A new similarity measure between signatures is proposed.
► We extract new data from iLIDS to evaluate multi-shot re-identification.