Adversarial attacks on artificial Intelligence of Things-based operational technologies in theme parks

Hina, Sadaf; Abbas, Qaiser; Ahmed, Kashan

Abstract

Theme parks represent a popular, yet vulnerable aspect of life, where large unsuspecting crowds gather and interact with technology. Artificial intelligence (AI), computer vision, and the Internet of Things (IoT) are transforming theme parks by revolutionizing various aspects. This research study is the first to identify critical components of theme parks that can be optimized, and comprehensively maps them onto emerging AI/IoT applications, often powered by machine learning or deep learning models. Additionally, the study sheds light on adversarial attacks targeting vulnerable smart surveillance systems, which generate a very large volume of video stream data. These systems serve as a prominent example of AIoT-based operational technologies (AIoT-OT) responsible for critical alerts and actions. Rigorous experimentation, involving a novel hybrid multi-pixel deception attack technique, demonstrates that advanced adversarial attack methods can significantly degrade the performance of detection systems. The performance metrics and attack success rate were measured by accuracy, precision, recall, F1-score, and AUC score. Before attack, the accuracy rates of 87. 45%, 83. 17% and 81. 40% were achieved for the EfficientNet, ResNet and MobileNet models, respectively. However, after applying the proposed MPD attack, the performance of each model declined significantly. The accuracy dropped to 61.23% for EfficientNet (with an attack success rate of 29.10%), 59.12% for ResNet (with success rate of 30.20%), and 55.17% for MobileNet (with success rate of 32.50%). This study signifies the need for a strategic plan of action and the development of robust methods for the proactive security of AIoT in theme parks.

• Journal Article Type: Article
• Acceptance Date: May 14, 2025
• Publication Date: 2025-05
• Deposit Date: Jun 4, 2025
• Publicly Available Date: Jun 9, 2025
• Journal: Internet of Things
• Print ISSN: 2542-6605
• Electronic ISSN: 2542-6605
• Publisher: Elsevier
• Peer Reviewed: Peer Reviewed
• Article Number: 101654
• Pages: 101654
• DOI: https://doi.org/10.1016/j.iot.2025.101654