View Invariant Appearance-Based Person Reidentification Using Fast Online Feature Selection and Score Level Fusion

Abstract

Fast and robust person reidentification is an important task in multi-camera surveillance and automated access control. We present an efficient appearance-based algorithm, able to reidentify a person regardless of occlusions, distance to the camera, and changes in view and lighting. The use of fast online feature selection techniques enables us to perform reidentification in hyper-real-time for a multicamera system, by taking only 10 seconds for evaluating 100 minutes of HD-video data. We demonstrate, that our approach surpasses current appearance-based state-of-theart in reidentification quality and computational speed and sets a new reference in non-biometric reidentification.

Viewpoint Invariant Human Re-Identification in Camera Networks Using Pose Priors and Subject-Discriminative Features

“…This is quite different from the case of person identification with high-definition videos [16], [27], [42].…”

“…There are also approaches to exploit discriminative or invariant features. Usually they are learned online, such as online feature selection [16], unsupervised salience learning [48], set-based methods [50] and covariance metric [2]. They can also be learned based on an offline learned dictionary, or using prototypes such as attribute-sensitive feature importance learning [31].…”

APFel: The intelligent video analysis and surveillance system for assisting human operators

“…For each comparison, a matching score is offered, based on the complement of the probability for a false acceptance, known as inverse logarithmic false acceptance rate score (or − log(FAR) score). For further details, it is referred to [5].…”

“…Therefore, it is important to use a large feature set for reidentification (e.g. texture features [9,13], color features and histograms [6]), and select discriminative features for a specific person on the fly in the enrollment phase [5]. Using a small subset of well suited features as a template ensures fast matching (12 000 per second).…”

“…For our system architecture, a fusion at score level is preferable. Therefore, the scheme of [5] is applied. We fuse the scores of the prediction module, face-and appearance-based recognition.…”