Improving vehicle re‐identification using CNN latent spaces: Metrics comparison and track‐to‐track extension

Geoffrey Roman‐Jimenez,Patrice Guyot,Sylvie Chambon,Alain Crouzil,Vincent Charvillat,André Péninou,Christine Sénac,Julien Pinquier,Florence Sedes,Thierry Malon

doi:10.1049/cvi2.12010

Abstract

This paper addresses the problem of vehicle re-identification using distance comparison of images in CNN latent spaces.Firstly, we study the impact of the distance metrics, comparing performances obtained with different metrics: the minimal Euclidean distance (MED), the minimal cosine distance (MCD), and the residue of the sparse coding reconstruction (RSCR). These metrics are applied using features extracted from five different CNN architectures, namely ResNet18, AlexNet, VGG16, InceptionV3 and DenseNet201. We use the specific vehicle re-identification dataset VeRi to fine-tune these CNNs and evaluate results. In overall, independently of the CNN used, MCD outperforms MED, commonly used in the literature. These results are confirmed on other vehicle retrieval datasets. Secondly, we extend the state-of-the-art image-to-track process (I2TP) to a track-to-track process (T2TP). The three distance metrics are extended to measure distance between tracks, enabling T2TP. We compared T2TP with I2TP using the same CNN models. Results show that T2TP outperforms I2TP for MCD and RSCR. T2TP combining DenseNet201 and MCD-based metrics exhibits the best performances, outperforming the state-of-the-art I2TP-based models. Finally, experiments highlight two main results: i) the impact of metric choice in vehicle re-identification, and ii) T2TP improves the performances compared to I2TP, especially when coupled with MCD-based metrics.

Highlights

With the recent growth of closed‐circuit television (CCTV) systems in big cities, object re‐identi"cation in video surveillance, such as vehicle and pedestrian re‐identi"cation, is a very active research "eld
Considering the performance improvement obtained with only global visual information of vehicle images and the very simplistic learning procedure that we used in our experiments ("ne‐tuning of standard CNN architectures), we argue that a relevant metric (MCD) combined with the use of more visual cues of the query vehicle (T2TP), could improve the performances of state‐of‐the‐art methods which are speci"cally designed for vehicle re‐ identi"cation
Recent studies on vehicle re‐identi"cation focused on the extraction of LR of vehicles, that is vectors of features extracted from the latent space of CNN, to discriminate between vehicles on their visual appearance to retrieve a given vehicle

Summary

Introduction

With the recent growth of closed‐circuit television (CCTV) systems in big cities, object re‐identi"cation in video surveillance, such as vehicle and pedestrian re‐identi"cation, is a very active research "eld. In the last few years, major progress has been observed in the vehicle re‐identi"cation "eld thanks to recent advances in machine‐ and deep‐learning [1]. These advances are very promising for intelligent video‐surveillance processing, intelligent transportation and future smart city systems. Vehicle re‐identi"cation, in video surveillance, aims at identifying a query vehicle, "lmed by one camera, among vehicles "lmed by other cameras of a CCTV system. It relies on a comparison between a query vehicle and a database of known vehicles, to "nd the best matches. The query is a single image and the vehicles of the database are represented by an image or a set of images called track, extracted from video segments recorded by CCTV cameras

Objectives

Methods

Results

Discussion

Conclusion