Swapping Faces, Saving Features: A Dual-Purpose Pipeline for Pedestrian Privacy in ITS

Jul 16, 2026

Large-scale, diverse pedestrian datasets are essential for training AI models used in autonomous vehicles within intelligent transportation systems (ITS), especially for pedestrian intention and trajectory prediction. However, unrestricted sharing of these images enables identity theft, tracking, and deepfake misuse. Prior privacy techniques often harm data usability by degrading facial attributes needed for training. This work introduces a five-stage face-swapping pipeline tailored for the EgyDRiVeS dataset, evaluating Roop and Ghost-v2, and using Roop to balance identity concealment with feature preservation.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages6

Reading Time15 min

Abstract

Introduction and Related Work

Swapping Faces, Saving Features: A Dual-Purpose Pipeline for Pedestrian Privacy in ITS Roba H. Farouk1 ,2 , and Catherine M. Elias 1 ,2 , Member, IEEE, arXiv :2607 .08402v 1 [ cs .CV] 9 Jul 2026 Abstract—Large-scale and diverse datasets are needed to train AI models to take real-time decisions for autonomous vehicles (AVs), an intelligent transportation system (ITS) application. Pedestrian intention and trajectory prediction are critical models used in AVs, requiring datasets involving diverse pedestrian images. Unrestricted access to these datasets imposes serious security risks, like identity theft and pedestrian tracking. The challenge is to apply privacy preservation procedures while maintaining the image attributes needed to train the models. Existing privacy methods may preserve the pedestrian’s privacy, but degrade the image usability, which hinders the models’ effectiveness. This work’s focus is to implement a five-stage pipeline to protect pedestrians’ privacy through face swapping while keeping the essential facial attributes intact. It should be tailored to satisfy the privacy needs of the EgyDRiVeS dataset. Moreover, Roop and Ghost-v2 face-swapping models are evaluated. Provenly, Roop outperforms Ghost-v2 in various aspects, as will be discussed. Consequently, Roop is the face-swapping model to be used in the pipeline to strike the balance between pedestrian privacy via identity concealment and data usability via facial attribute preservation. Index Terms—Pedestrian Privacy, Data Usability, Datasets, Face Swapping, Features Preservation. I. INTRODUCTION AND RELATED WORK ITS has contributed to the mobility industry by making systems smarter, safer, and more dynamic. To empower ITS, thriving technologies including AI, global positioning systems (GPS), the internet of things (IoT), and big data must be exploited. Some AI models allow for decision-making abilities and perception mechanisms to be used in AVs, such as pedestrian intention and trajectory prediction and path planning. To train these models, large-scale and diverse datasets are a must. Using AI-based models, the vehicles should be able to predict pedestrians’ anticipated behavior and take appropriate real-time decisions that avoid potential collisions and ensure pedestrians’ safety. Datasets including real pedestrian footage to study their natural behavior in various situations are indispensable. Crucially, facial cuesand body movements are the main features needed by these models [1]–[3] . The facial region is in critical need of protection, as it reveals identifiable biometric data about the pedestrian. Unauthorized access to these datasets imposes serious security threats on pedestrians, including identity theft, personal information extraction through data mining, surveillance tracking, and deepfake generation. *This work was not supported by any organization 1C-DRiVeS Lab: Cognitive Driving Research in Vehicular Systems, Cairo, Egypt [cdrives.researchlab@gmail.com](cdrives.researchlab@gmail.com) [2](2 Computer Science and Engineering Department - Faculty of Media)[ Computer Science and Engineering Department - Faculty of Media](2 Computer Science and Engineering Department - Faculty of Media)[ ](2 Computer Science and Engineering Department - Faculty of Media)Engineering and Technology -German University in Cairo, Egypt [roba.ali@student.guc.edu.eg](roba.ali@student.guc.edu.eg) , [catherine.elias@ieee.org](catherine.elias@ieee.org) Some countries have established global privacy regulations, such as the General Data Protection Regulation (GDPR) in Europe and the California Consumer Privacy Act (CCPA) . Some developing countries are building their own urban datasets while complying with the legal and ethical privacy preservation guidelines. The Egy-DRiVeS dataset is an Egyptian dataset that includes images and videos taken from the public streets [4] . These urban datasets need specific privacy preservation procedures to handle th

FAQs

Why are pedestrian image datasets important for ITS and autonomous vehicles?

They train models for real-time pedestrian intention and trajectory prediction, enabling safer path planning and collision avoidance in autonomous vehicles.

What security risks arise from unrestricted access to pedestrian datasets?

They can lead to identity theft, personal information extraction, surveillance tracking, and potential deepfake generation.

How does the proposed pipeline balance privacy and data usability?

It uses a five-stage face-swapping approach that obscures pedestrian identity while preserving essential facial attributes required for effective model training, with Roop selected over Ghost-v2.

Jul 16, 2026

Upload by Valentina