Pose-Driven Deep Models for Person Re-Identification

TL;DR

This paper introduces a CNN-based person re-identification method that incorporates camera view and pose information directly into the model, significantly improving robustness and accuracy across multiple datasets.

Contribution

It demonstrates that integrating coarse camera view and fine-grained pose data directly into CNNs enhances person re-id performance without complex pre-processing.

Findings

Significant improvements over state-of-the-art on four datasets

Effective use of camera view and pose info within CNNs

Introduction of X-MARS dataset for cross-validation

Abstract

Person re-identification (re-id) is the task of recognizing and matching persons at different locations recorded by cameras with non-overlapping views. One of the main challenges of re-id is the large variance in person poses and camera angles since neither of them can be influenced by the re-id system. In this work, an effective approach to integrate coarse camera view information as well as fine-grained pose information into a convolutional neural network (CNN) model for learning discriminative re-id embeddings is introduced. In most recent work pose information is either explicitly modeled within the re-id system or explicitly used for pre-processing, for example by pose-normalizing person images. In contrast, the proposed approach shows that a direct use of camera view as well as the detected body joint locations into a standard CNN can be used to significantly improve the robustness of learned re-id embeddings. On four challenging surveillance and video re-id datasets significant improvements over the current state of the art have been achieved. Furthermore, a novel reordering of the MARS dataset, called X-MARS is introduced to allow cross-validation of models trained for single-image re-id on tracklet data.