Modeling feature distances by orientation driven classifiers for person re-identification

Abstract

To tackle the re-identification challenges existing methods propose to directly match image features or to learn the transformation of features that undergoes between two cameras. Other methods learn optimal similarity measures. However, the performance of all these methods are strongly dependent from the person pose and orientation. We focus on this aspect and introduce three main contributions to the field: (i) to propose a method to extract multiple frames of the same person with different orientations in order to capture the complete person appearance; (ii) to learn the pairwise feature dissimilarities space (PFDS) formed by the subspaces of similar and different image pair orientations; and (iii) within each subspace, a classifier is trained to capture the multi-modal inter-camera transformation of pairwise image dissimilarities and to discriminate between positive and negative pairs. The experiments show the superior performance of the proposed approach with respect to state-of-the-art methods using two publicly available benchmark datasets. © 2016 Elsevier Inc. All rights reserved

Re-ranking Person Re-identification with k-Reciprocal Encoding

“…Re-ranking for re-ID. Most existing person reidentification methods mainly focus on feature representation [43,13,24,50,22] or metric learning [24,18,10,33,47]. Recently, several researchers [11,34,29,51,21,12,20,44,46] have paid attention to re-ranking based method in the re-ID community.…”

Robust and Efficient Person Re-Identification Model using K-Nearest Neighbor Graph

“…Moreover, an ideal distance metric also plays as an important tool for the Hand Crafted feature but the metric learning might not work in case of the deep feature and the main reason behind this is it has learned sufficient information for the robust feature of that particular person and for the detailed [11] and [16] can be referred. In past several person identification technique has been proposed these have mainly focused on the metric learning [19] or the feature learning [18], other technique as discussed in [20][21][22][23][24] have focused on the re-ranking technique, for the re-identification and it totally differs from paper [25] and [26]. These two requires either label supervision or human interaction and it focus on the unsupervised and automatic solution.…”

Group Re-identification via Unsupervised Transfer of Sparse Features Encoding

“…In particular, metric learning approaches have been proposed by relaxing [19] or enforcing [30] positive semi-definite (PSD) conditions, by considering equivalence constraints [26,47,46] or by exploiting the null-space [53]. While most of the existing methods capture the global structure of the dissimilarity space, local solutions [28,41,13] have been proposed too. Sample-specific metrics were also investigated in [54].…”

“…In particular, existing works have focused on seeking either the best feature representations (e.g., [31,39,29]) or propose to learn optimal matching metrics (e.g., [26,30,53]). Despite obtaining interesting results on benchmark datasets (e.g., [13,40,61], such works have generally neglected the fact that in crowded public environments people often walk in groups.…”