Optical Lens Attack on Monocular Depth Estimation for Autonomous Driving

Ce Zhou,Qiben Yan,Daniel Kent,Guangjing Wang,Weikang Ding,Ziqi Zhang,Hayder Radha

2024-11-01

Abstract:Monocular Depth Estimation (MDE) is a pivotal component of vision-based Autonomous Driving (AD) systems, enabling vehicles to estimate the depth of surrounding objects using a single camera image. This estimation guides essential driving decisions, such as braking before an obstacle or changing lanes to avoid collisions. In this paper, we explore vulnerabilities of MDE algorithms in AD systems, presenting LensAttack, a novel physical attack that strategically places optical lenses on the camera of an autonomous vehicle to manipulate the perceived object depths. LensAttack encompasses two attack formats: concave lens attack and convex lens attack, each utilizing different optical lenses to induce false depth perception. We first develop a mathematical model that outlines the parameters of the attack, followed by simulations and real-world evaluations to assess its efficacy on state-of-the-art MDE models. Additionally, we adopt an attack optimization method to further enhance the attack success rate by optimizing the attack focal length. To better evaluate the implications of LensAttack on AD, we conduct comprehensive end-to-end system simulations using the CARLA platform. The results reveal that LensAttack can significantly disrupt the depth estimation processes in AD systems, posing a serious threat to their reliability and safety. Finally, we discuss some potential defense methods to mitigate the effects of the proposed attack.

Computer Vision and Pattern Recognition,Cryptography and Security

What problem does this paper attempt to address?

### What problems does this paper attempt to solve?

This paper aims to explore and reveal the potential vulnerabilities of the Monocular Depth Estimation (MDE) algorithm in the autonomous driving system, and proposes a new physical attack method named **𝐿𝑒𝑛𝑠𝐴𝑡𝑡𝑎𝑐𝑘**. Specifically, this research strategically places optical lenses on the cameras of autonomous vehicles to manipulate the perceived object depth, thereby interfering with the normal operation of the autonomous driving system.

#### Main problem description:
1. **Vulnerability of the MDE algorithm**: MDE is a key component in vision - based autonomous driving systems. It estimates the depth of surrounding objects through a single - camera image and provides guidance for driving decisions such as braking and lane changing. However, these algorithms are vulnerable to physical attacks.

2. **Innovation of attack means**: Previous studies have mainly focused on adversarial attacks in the digital world or visible adversarial patches, but these methods have limited effectiveness in the real world. Therefore, this paper proposes a new method of physical attack using optical lenses, namely **𝐿𝑒𝑛𝑠𝐴𝑡𝑡𝑎𝑐𝑘**.

3. **Specific forms of the attack**: The attack is divided into two forms:
- **Concave lens attack**: By placing a concave lens, the object appears farther away.
- **Convex lens attack**: By placing a convex lens, the object appears closer.

4. **Effectiveness and optimization of the attack**: To ensure the success rate of the attack, the author has developed a mathematical model to describe the attack parameters and evaluated its effect through simulation and actual tests. In addition, an optimization method is introduced to adjust the focal length of the attack to further improve the success rate of the attack.

5. **System - level impact assessment**: To comprehensively assess the impact of **𝐿𝑒𝑛𝑠𝐴𝑡𝑡𝑎𝑐𝑘** on the autonomous driving system, the author uses the CARLA platform for end - to - end system simulation, demonstrating the potential threat of this attack in the real world.

#### Conclusion:
The experimental results show that **𝐿𝑒𝑛𝑠𝐴𝑡𝑡𝑎𝑐𝑘** can significantly disrupt the depth estimation process in the autonomous driving system, posing a serious threat to the reliability and safety of the system. At the same time, the author also discusses some possible defense methods to mitigate the impact of this attack.

### Summary of mathematical formulas
The formulas involved in the paper mainly include the basic formulas for lens imaging:

1. **Lens formula**:
\[
\frac{1}{f}=\frac{1}{d_o}+\frac{1}{d_i}
\]
where \( f \) is the focal length, \( d_o \) is the object distance, and \( d_i \) is the image distance.

2. **Magnification formula**:
\[
m =-\frac{d_i}{d_o}
\]
where \( m \) is the magnification.

3. **Total magnification formula**:
\[
m_{\text{total}}=m_1m_2=\frac{f f_c}{(d_{o1}-f)(|d_{o1}f/(d_{o1}-f)|+d_b - f_c)}
\]
where \( f_c \) is the focal length of the camera lens, and \( d_b \) is the distance between the attack lens and the camera lens.

These formulas are used to describe the influence of the lens on the image size and depth estimation during the attack process.