Re-Identification of a person in the context of video surveillance systems is of great interest in several applications such as the generation of consistent trajectories of people across widespread fields of view of cameras, reidentification of a person temporally not visible to the system, or the search for a certain person in video footage. In this paper we address the problem of person re-identification in complex non-cooperative environments - those with low resolution and challenging lighting conditions. We address the problem in a threefold manner: first we show how local visual features, extracted on a person's image can be used for robust person tracking and appearance based re-identification. Second, based on the person detection and tracking, face-based features are extracted. Hereby, super-resolution techniques are used for face image improvement at a distance. Finally, extraction of additional soft-biometric semantic features for re-identification will be discussed. All approaches towards re-identification are presently examined under the research project Remote Biometrics ("Distante Biometrie") of the Fraunhofer Institute of Optronics, System Technologies, and Image Exploitation (IOSB).
展开▼
