Person Re-Identification by Discriminative Selection in Video Ranking

Abstract

Current person re-identification (ReID) methods typically rely on single-frame imagery features, whilst ignoring space-time information from image sequences often available in the practical surveillance scenarios. Single-frame (single-shot) based visual appearance matching is inherently limited for person ReID in public spaces due to the challenging visual ambiguity and uncertainty arising from non-overlapping camera views where viewing condition changes can cause significant people appearance variations. In this work, we present a novel model to automatically select the most discriminative video fragments from noisy/incomplete image sequences of people from which reliable space-time and appearance features can be computed, whilst simultaneously learning a video ranking function for person ReID. Using the PRID 2011, iLIDS-VID, and HDA+ image sequence datasets, we extensively conducted comparative evaluations to demonstrate the advantages of the proposed model over contemporary gait recognition, holistic image sequence matching and state-of-the-art single-/multi-shot ReID methods.

Intra-Camera Supervised Person Re-Identification: A New Benchmark

“…Supervised person re-id Most existing person re-id models are created by supervised learning methods on a separate set of cross-camera identity labelled training data (Wang et al, 2014b(Wang et al, , 2016bZhao et al, 2017;Chen et al, 2017;Li et al, 2017;Chen et al, 2018b;Li et al, 2018b;Song et al, 2018;Chang et al, 2018;Sun et al, 2018;Shen et al, 2018a;Wei et al, 2018;Hou et al, 2019;Zheng et al, 2019;Zhang et al, 2019;Quan et al, 2019;Zhou et al, 2019). Relying on the strong supervision of cross-camera identity labelled training data, they have achieved remarkable performance boost.…”

GAN-Based Pose-Aware Regulation for Video-Based Person Re-Identification

“…The periodicity of pedestrians gate is exploited also in [24] to generate a spatio-temporal bodyaction model made up of a series of action primitives of certain body parts treated independently from each other. The strict alignment assumptions made by gait recognition techniques are relaxed in [47,61] where unregulated videosequences are automatically broken down based on motion energy profiling (e.g., optical flow). With the spreading of the deep learning paradigm, temporal-based approaches have emerged using RNNs (combined with Convolutional Neural Networks) in Siamese configuration [30,56,52,71].…”

“…For the video-based person re-id, given a query sequence of images (a.k.a., tracklet) of the target of interest, the challenge consists of identifying all the corresponding matching tracklets captured across the network. Dealing with full sequences of images as opposed to single images offers significant advantages: a) exploiting the temporal dependencies between intra-sequence frames [52,30,56,71]; b) extracting more robust spatial appearance descriptors [64,51,58]; c) partially recovering from occlusions and reducing the influence of the background [46,47,13]. All these three aspects contribute to reducing the impact of the factors affecting performance (changing pose/viewpoint, lighting, occlusions, etc.…”

“…Temporal alignment methods usually suffer from noisy unregulated sequences with severe viewpoint and lighting variability [47]. Under such scenarios, it is difficult to estimate accurately gait phase and cycle or the optical flow energy profile as in [61].…”

Bifocal Stereo for Multipath Person Re-Identification

“…For test and training data ten random splits (cf. (Gray et al, 2007)) of often cited datasets VIPeR (Gray et al, 2007), prid4502 (Roth et al, 2014), and i-lids (Wang et al, 2016) (Wang et al, 2014) were analysed. The same splits are used for every down-sampled trial.…”