The Research of Group Re-identification from Multiple Cameras

TL;DR

This work tackles group re-identification across multiple cameras by introducing multi-granularity representations of groups (finer: individuals; middle: two-person subgroups; coarse: three-person subgroups) and an importance-driven framework that iteratively weighs each object's contribution using saliency, purity, and stability. It combines a multi-order matching approach on a multi-order association graph with representative selection to fuse information across granularities, enabling robust one-to-one group matching despite layout changes and varying group membership. Datasets including MCTS, DukeMTMC, and a new Road dataset demonstrate that multi-granularity features and multi-order matching yield superior accuracy over baselines and prior group Re-ID methods, with iterative refinement converging within a few steps. The framework highlights practical implications for surveillance and tracking, offering a scalable approach to identify groups across non-overlapping camera views, while outlining avenues for improved detection integration, convergence analysis, and computational efficiency.

Abstract

Object re-identification is of increasing importance in visual surveillance. Most existing works focus on re-identify individual from multiple cameras while the application of group re-identification (Re-ID) is rarely discussed. We redefine Group Re-identification as a process which includes pedestrian detection, feature extraction, graph model construction, and graph matching. Group re-identification is very challenging since it is not only interfered by view-point and human pose variations in the traditional re-identification tasks, but also suffered from the challenges in group layout change and group member variation. To address the above challenges, this paper introduces a novel approach which leverages the multi-granularity information inside groups to facilitate group re-identification. We first introduce a multi-granularity Re-ID process, which derives features for multi-granularity objects (people/people-subgroups) in a group and iteratively evaluates their importances during group Re-ID, so as to handle group-wise misalignments due to viewpoint change and group dynamics. We further introduce a multi-order matching scheme. It adaptively selects representative people/people-subgroups in each group and integrates the multi-granularity information from these people/people-subgroups to obtain group-wise matching, hence achieving a more reliable matching score between groups. Experimental results on various datasets demonstrate the effectiveness of our approach.

Figures (13)

• Figure 1: Vitual scenarios of multiple cameras surveillance
• Figure 2: Example of some challenging true group pair. (a): A true group pair with large layout variation. (b): A true group pair with different object number. (c): A true group pair with large viewpoint change. (Best viewed)
• Figure 3: (a) 2 visually similar gallery image in camera B to the same probe image in camera A. The true pair will be easily confused if we only consider individual or global object. If we consider middle-level group granularity of two-people subgroup and assign different importance of each people/people-subgroups, we are able to obtain a more accurate group re-identification result. (The red color denotes high value of importance and the blue color shows low value of importance) (b) Illustration of different importance of multi-granularity. (Best viewed in color)
• Figure 4: Framework of the proposed approach.
• Figure 5: Framework of pedestrian detection reinspect