K. T. Yasas Mahima
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
K. T. Yasas Mahima has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Asanka G. Perera
School of Engineering & Digital Technologies, University of Southern Queensland, Springfield Central, QLD 4300, Australia
4 shared publications - Matthew Garratt
School of Engineering and Technology, University of New South Wales, Canberra, ACT 2612, Australia
3 shared publications - Sreenatha G. Anavatti
School of Engineering and Technology, University of New South Wales, Canberra, ACT 2612, Australia
3 shared publications - Sandesh Shrestha
Industrial Systems Engineering (ISE), Asian Institute of Technology (AIT), Pathum Thani 12120, Thailand
1 shared publication
Latest Publications
- Open Access
Adversarial Patch and Camouflage Attacks on Aerial Object Detection: A Geometry-Aware Review
Uncrewed aerial vehicles increasingly rely on on-board deep object detectors for applications such as surveillance, delivery, agriculture, and traffic monitoring. The vulnerability of these detectors to adversarial patches has therefore become a practical safety concern. Research on adversarial patch and camouflage attacks against aerial object detection has expanded rapidly, yet existing reviews consider aerial systems only as one application domain among many and do not examine the unique geometric conditions of aerial imagery. This review surveys adversarial patch and camouflage attacks, together with the corresponding defense mechanisms, for aerial, UAV, and remote sensing object detection. The literature is organized using a taxonomy based on two dimensions: the physical medium of the perturbation and the application domain it targets. Existing methods are then compared with respect to physical robustness, evaluation datasets and detectors, threat models, and their treatment of viewpoint variation. A recurring observation across the literature is that viewpoint variation is represented primarily through object scale, while explicit modeling of viewing geometry remains limited. This trend is closely linked to the characteristics of the aerial datasets used for evaluation, which typically provide realistic imagery but limited geometric information. Consequently, questions concerning viewpoint-dependent placement, degradation, and naturalness remain only partially explored. The review concludes by identifying the major research gaps, future directions, and deployment challenges that are likely to shape the next generation of aerial adversarial attack and defense research.
Open AccessAdversarial Patch and Camouflage Attacks on Aerial Object Detection: A Geometry-Aware Review
Uncrewed aerial vehicles increasingly rely on on-board deep object detectors for applications such as surveillance, delivery, agriculture, and traffic monitoring. The vulnerability of these detectors to adversarial patches has therefore become a practical safety concern. Research on adversarial patch and camouflage attacks against aerial object detection has expanded rapidly, yet existing reviews consider aerial systems only as one application domain among many and do not examine the unique geometric conditions of aerial imagery. This review surveys adversarial patch and camouflage attacks, together with the corresponding defense mechanisms, for aerial, UAV, and remote sensing object detection. The literature is organized using a taxonomy based on two dimensions: the physical medium of the perturbation and the application domain it targets. Existing methods are then compared with respect to physical robustness, evaluation datasets and detectors, threat models, and their treatment of viewpoint variation. A recurring observation across the literature is that viewpoint variation is represented primarily through object scale, while explicit modeling of viewing geometry remains limited. This trend is closely linked to the characteristics of the aerial datasets used for evaluation, which typically provide realistic imagery but limited geometric information. Consequently, questions concerning viewpoint-dependent placement, degradation, and naturalness remain only partially explored. The review concludes by identifying the major research gaps, future directions, and deployment challenges that are likely to shape the next generation of aerial adversarial attack and defense research.
- Open Access
- 4 March 2025
- Published by Elsevier in Pattern Recognition Letters
- Vol. 191, 37-43
FlowCraft: Unveiling adversarial robustness of LiDAR scene flow estimation
- 4 March 2025
- Published by Elsevier in Pattern Recognition Letters
- Vol. 191, 37-43
Open AccessFlowCraft: Unveiling adversarial robustness of LiDAR scene flow estimation
- 17 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 25 (12) , 19176-19202
Toward Robust 3D Perception for Autonomous Vehicles: A Review of Adversarial Attacks and Countermeasures
At present the perception system of autonomous vehicles is grounded on 3D vision technologies along with deep learning to process depth information. Although deep learning models for 3D perception give promising results, recent research demonstrates that they are also vulnerable to adversarial attacks similar to deep learning models trained on 2D images. As a result, it is essential to further explore the vulnerabilities of 3D perception models in autonomous vehicles and find methods to cope with the risks associated with these adversarial vulnerabilities, in order to improve the social acceptance of commercial autonomous vehicles. This study aims to provide an in-depth overview of the recent adversarial attacks and countermeasures against 3D perception models on autonomous vehicles. Further, challenges associated with the research domain and future research directions are highlighted to make autonomous vehicles robust against adversarial attacks.
- 17 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 25 (12) , 19176-19202
Toward Robust 3D Perception for Autonomous Vehicles: A Review of Adversarial Attacks and Countermeasures
At present the perception system of autonomous vehicles is grounded on 3D vision technologies along with deep learning to process depth information. Although deep learning models for 3D perception give promising results, recent research demonstrates that they are also vulnerable to adversarial attacks similar to deep learning models trained on 2D images. As a result, it is essential to further explore the vulnerabilities of 3D perception models in autonomous vehicles and find methods to cope with the risks associated with these adversarial vulnerabilities, in order to improve the social acceptance of commercial autonomous vehicles. This study aims to provide an in-depth overview of the recent adversarial attacks and countermeasures against 3D perception models on autonomous vehicles. Further, challenges associated with the research domain and future research directions are highlighted to make autonomous vehicles robust against adversarial attacks.
- Open Access
Exploring Adversarial Robustness of LiDAR Semantic Segmentation in Autonomous Driving
Deep learning networks have demonstrated outstanding performance in 2D and 3D vision tasks. However, recent research demonstrated that these networks result in failures when imperceptible perturbations are added to the input known as adversarial attacks. This phenomenon has recently received increased interest in the field of autonomous vehicles and has been extensively researched on 2D image-based perception tasks and 3D object detection. However, the adversarial robustness of 3D LiDAR semantic segmentation in autonomous vehicles is a relatively unexplored topic. This study expands the adversarial examples to LiDAR-based 3D semantic segmentation. We developed and analyzed three LiDAR point-based adversarial attack methods on different networks developed on the SemanticKITTI dataset. The findings illustrate that the Cylinder3D network has the highest adversarial susceptibility to the analyzed attacks. We investigated how the class-wise point distribution influences the adversarial robustness of each class in the SemanticKITTI dataset and discovered that ground-level points are extremely vulnerable to point perturbation attacks. Further, the transferability of each attack strategy was assessed, and we found that networks relying on point data representation demonstrate a notable level of resistance. Our findings will enable future research in developing more complex and specific adversarial attacks against LiDAR segmentation and countermeasures against adversarial attacks.
Open AccessExploring Adversarial Robustness of LiDAR Semantic Segmentation in Autonomous Driving
Deep learning networks have demonstrated outstanding performance in 2D and 3D vision tasks. However, recent research demonstrated that these networks result in failures when imperceptible perturbations are added to the input known as adversarial attacks. This phenomenon has recently received increased interest in the field of autonomous vehicles and has been extensively researched on 2D image-based perception tasks and 3D object detection. However, the adversarial robustness of 3D LiDAR semantic segmentation in autonomous vehicles is a relatively unexplored topic. This study expands the adversarial examples to LiDAR-based 3D semantic segmentation. We developed and analyzed three LiDAR point-based adversarial attack methods on different networks developed on the SemanticKITTI dataset. The findings illustrate that the Cylinder3D network has the highest adversarial susceptibility to the analyzed attacks. We investigated how the class-wise point distribution influences the adversarial robustness of each class in the SemanticKITTI dataset and discovered that ground-level points are extremely vulnerable to point perturbation attacks. Further, the transferability of each attack strategy was assessed, and we found that networks relying on point data representation demonstrate a notable level of resistance. Our findings will enable future research in developing more complex and specific adversarial attacks against LiDAR segmentation and countermeasures against adversarial attacks.