Optimizing vehicle re-identification: extended neighbourhood distance re-ranking for images captured from multi-camera surveillance systems

With the growth of smart cities, vast research has been performed on vehicle re-identification for intelligent transportation. Vehicle re-identification (re-ID) involves gathering accurate vehicle information from multiple disjoint surveillance cameras. The paper presents an advanced feature extract...

Full description

Saved in:
Bibliographic Details
Main Authors: B. Cynthia Sherin, E. Poovammal
Format: Article
Language:English
Published: Taylor & Francis Group 2025-10-01
Series:Automatika
Subjects:
Online Access:https://www.tandfonline.com/doi/10.1080/00051144.2025.2540164
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:With the growth of smart cities, vast research has been performed on vehicle re-identification for intelligent transportation. Vehicle re-identification (re-ID) involves gathering accurate vehicle information from multiple disjoint surveillance cameras. The paper presents an advanced feature extraction framework called FastRe-ID, which uses an attention-augmented residual network (Att-ResNet50) for object re-identification. This attention-guided network extracts the vehicle features from the images for precise identification. To further improve the re-identification performance, the suggested work leverages a novel re-ranking technique based on extended neighbourhood distance (END). When two vehicles captured in different images appear identical, their respective initial rank lists and two-level neighbourhood images also remain identical. The proposed method overcomes these similarity issues by incorporating a unique END distance generated from the extended neighbours along with the initial rank list. The END distance is aggregated along with the Jaccard distance to obtain the final distance and produce accurate images in the final rank list. Our proposed re-ranking method has accomplished a 10.13% increase on the VeRi-776 and 5.96% on the VeRi-Wild dataset in accuracy, and 4.44%, 4.51%, and 5.31% increases in accuracy on the small, medium, and large sets of the Vehicle-ID dataset, respectively.
ISSN:0005-1144
1848-3380