Weighted Average Precision: Adversarial Example Detection in the Visual Perception of Autonomous Vehicles

Li, Yilan; Velipasalar, Senem

doi:10.48550/arxiv.2002.03751

Cited by 2 publications

(5 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We achieved the highest detection rate of 97.4 while detecting shot-noised adversarial attacks, while the lowest detection rate of 90% was recorded when detecting Gaussian-noise-based adversarial attacks. On the other hand, the highest detection rate of the detector proposed in [79] was recorded at 74.1% in detecting FGSM attacks in the black-box setting.…”

Section: Comparative Analysis With Existing Approachesmentioning

confidence: 95%

“…We compared attacking models, the approach to detect adversarial attacks, and the performance of the proposed approach with [78,79]. We considered the most commonly used black-box attacks to attack the target model because white-box attacks are hard to implement in ADSs in real time.…”

Section: Comparative Analysis With Existing Approachesmentioning

confidence: 99%

“…Our deep-autoencoder-based attack detector achieved a high detection rate in detecting black-box attacks. The detection accuracy of our proposed method was compared with [79] in the black-box setting. Compared to [79], our approach achieved an average of 94% detection rate while detecting black-box-type adversarial attacks.…”

Section: Comparative Analysis With Existing Approachesmentioning

confidence: 99%

“…The detection accuracy of our proposed method was compared with [79] in the black-box setting. Compared to [79], our approach achieved an average of 94% detection rate while detecting black-box-type adversarial attacks. We achieved the highest detection rate of 97.4 while detecting shot-noised adversarial attacks, while the lowest detection rate of 90% was recorded when detecting Gaussian-noise-based adversarial attacks.…”

Section: Comparative Analysis With Existing Approachesmentioning

confidence: 99%

See 3 more Smart Citations

Reconstruction-Based Adversarial Attack Detection in Vision-Based Autonomous Driving Systems

Hussain,

Hong

2023

MAKE

View full text Add to dashboard Cite

The perception system is a safety-critical component that directly impacts the overall safety of autonomous driving systems (ADSs). It is imperative to ensure the robustness of the deep-learning model used in the perception system. However, studies have shown that these models are highly vulnerable to the adversarial perturbation of input data. The existing works mainly focused on studying the impact of these adversarial attacks on classification rather than regression models. Therefore, this paper first introduces two generalized methods for perturbation-based attacks: (1) We used naturally occurring noises to create perturbations in the input data. (2) We introduce a modified square, HopSkipJump, and decision-based/boundary attack to attack the regression models used in ADSs. Then, we propose a deep-autoencoder-based adversarial attack detector. In addition to offline evaluation metrics (e.g., F1 score and precision, etc.), we introduce an online evaluation framework to evaluate the robustness of the model under attack. The framework considers the reconstruction loss of the deep autoencoder that validates the robustness of the models under attack in an end-to-end fashion at runtime. Our experimental results showed that the proposed adversarial attack detector could detect square, HopSkipJump, and decision-based/boundary attacks with a true positive rate (TPR) of 93%.

show abstract

Section: Comparative Analysis With Existing Approachesmentioning

confidence: 95%

Section: Comparative Analysis With Existing Approachesmentioning

confidence: 99%

Section: Comparative Analysis With Existing Approachesmentioning

confidence: 99%

Section: Comparative Analysis With Existing Approachesmentioning

confidence: 99%

See 2 more Smart Citations

Reconstruction-Based Adversarial Attack Detection in Vision-Based Autonomous Driving Systems

Hussain,

Hong

2023

MAKE

View full text Add to dashboard Cite

show abstract

“…The average AP over various IoU thresholds is called mAP. Average AP over all classes [27]. The formula of mAP is as in (5).…”

Section: Mean Average Precision (Map)mentioning

confidence: 99%

An Efficient and Fast Lightweight-Model with ShuffleNetv2 Based on YOLOv5 for Detection of Hardhat-Wearing

Cengil¹,

Çınar²,

Yıldırım³

2022

RCES

View full text Add to dashboard Cite

Traumatic brain injuries and collisions from falls and electric shocks are among the leading causes of construction deaths. Helmets play an important role in protecting working people from accidents. However, wearing a hard hat in real life is often not strictly enforced among those who try. Therefore, it is important to check this and ensure that a helmet is worn. Today, the use of artificial intelligence-based object recognition systems has become widespread due to the advantages it provides. In this article, a one-step object detection approach based on deep learning is proposed to detect helmet use and control helmet wearing status. The model is based on the YOLOv5 architecture. In the feature extraction step of the method, ShuffleNetv2, which is a lightweight model for a fast detector, is used. The presented model has been examined on the Hard Hat Workers dataset. The architecture provided a recall value of 0.942 precision 0.91 in the corresponding dataset. The obtained results showed that the recommended model is suitable for use on construction sites to check whether a helmet is fitted.

show abstract

Weighted Average Precision: Adversarial Example Detection in the Visual Perception of Autonomous Vehicles

Cited by 2 publications

References 42 publications

Reconstruction-Based Adversarial Attack Detection in Vision-Based Autonomous Driving Systems

Reconstruction-Based Adversarial Attack Detection in Vision-Based Autonomous Driving Systems

An Efficient and Fast Lightweight-Model with ShuffleNetv2 Based on YOLOv5 for Detection of Hardhat-Wearing

Contact Info

Product

Resources

About