Stationary Case Studies

Experimental Setup

For these case studies, the attacker and victim radar were placed 15m away from each other. At the start of the case study, a target vehicle begins to drive away from the victim radar at a velocity of roughly 20mph. (~9m/s).

When reviewing the following results, we make the following notes:

The minimum detection range of the victim radar is 24.35 m*.
Detections corresponding to stationary vehicles and objects are not shown in the plots below**.

Stationary Case Studies

Whether it is an infrastructure sensor detecting vehicles at a stoplight or a stopped vehicle sensing oncoming traffic prior to pulling out of a parking lot, accurately detecting one's surroundings is critical to safety on the roads. To showcase how a malicious actor might compromise such sensors, we present the following case studies where a stationary attacker is set up to attack a stationary victim. Here, the victim attempts to detect objects on the road. Meanwhile, the attacker estimates the victim's settings in real-time and then simultaneously launches various attacks designed to impede the victim's ability to accurately sense vehicles in the environment. The figure on the right illustrates the setup we used to perform these case studies. To the best of our knowledge, we are the first to present the following four black-box attacks using a real-time prototype system. Use the links below to explore our case studies.

No Attack

Multiple False Positive Attacks

False Negative Attack

Translation Attack

No Attack.mp4

Case Study Movie

No Attack

Here we demonstrate the baseline performance of the victim radar detecting objects moving along the road. This demonstrates the radar's baseline performance when no vehicle is present.

Detected Object Locations

This figure documents the range that the victim radar perceived the target to be for each frame of the case study. When not under attack, the victim radar detects the target moving away from it in a majority of its frames

Attack Timeline

The figure above documents the progression of the case study. Here, the top row shows the Range-Doppler response while the bottom row shows the CA-CFAR detections. Observe how the radar successfully detects the black car moving away from it for the duration of the case study.

Multiple FP Attacks.mp4

Attack Movie

Multiple False Positive Attacks

In this case study, we add two additional fake objects to the victim's radar point cloud. The first object starts at a range of 80m and then propagates toward the radar with a velocity of 10m/s. Simultaneously, a second spoofed object starts at a range of 40m and propagates away from the radar with a velocity of 10 m/s. Here, observe how the added points behave like real objects where the reflected power dynamically increases or decreases as the objects get closer or further away respectively.

Detected Object Locations

This figure documents the range that the victim radar perceived the target to be for each frame of the case study. Observe how the radar successfully detects both of the real objects (the black car and the white car). However, we highlight how the radar also detects the two additional spoofed objects as well.

Attack Timeline

The figure above documents the progression of the case study. Here, the top row shows the Range-Doppler response while the bottom row shows the CA-CFAR detections. The first column shows the radar's perception when not under attack while the second and third columns represent the radar's perception when under attack. Observe how hen no attack is present, we clearly see the black car being detected moving away from the radar when no attack is present. However, when the attack is engaged, we see that the victim also detects both of the false-positive attacks.

FN Attack.mp4

Attack Movie

False Negative Attack

In this case study, we configure our attacker to apply a false negative attack centered at a range of 75m with a velocity of -10 m/s. Here, observe how this added clutter raises the CFAR threshold of the victim's radar to the point that the real object is not detected once the attack starts.

Detected Object Locations

This figure documents the range that the victim radar perceived the target to be for each frame of the case study. Here, the False-Negative attack starts on the 14th frame where we observe that the victim radar fails to detect the object once the attack starts.

Attack Timeline

The figure above documents the progression of the case study. Here, the top row shows the Range-Doppler response while the bottom row shows the CA-CFAR detections. The first column shows the radar's perception when not under attack while the second and third columns represent the radar's perception when under attack. When no attack is present, we clearly see the black car moving away from the radar in the Range-Doppler response where it is also detected using the CFAR. However, when the attack is engaged, observe how the car is no longer detected by the CFAR detector even though it is still present in the scene.

Translation Attack.mp4

Attack Movie

Translation Attack

In this case study, we combine a false-negative attack centered at a range of 75m and a velocity of -10m/s with a false positive attack starting at 75m and propagating towards the radar with a velocity of 12m/s. The resulting attack leads the victim radar to mistakenly believe that the car is heading towards it when it is actually heading away from it.

Detected Object Locations

This figure documents the range that the victim radar perceived the target to be for each frame of the case study. Prior to the attack, we see that the victim successfully tracks the vehicle as it moves away from it. However, observe how the radar mistakenly believes that the object is instead coming towards it once the translation attack starts.

Attack Timeline

The figure above documents the progression of the case study. Here, the top row shows the Range-Doppler response while the bottom row shows the CA-CFAR detections. The first column shows the radar's perception when not under attack while the second and third columns represent the radar's perception when under attack. Observe how we clearly see the bus moving away from the radar in the Range-Doppler response where it is also detected using the CFAR when no attack is present. However, when the attack is engaged, observe how the car is mistakenly detected as coming toward the radar instead of moving away from it.

Footnotes:

*This is a result of the CA-CFAR detector configuration (see Sec.VII in the paper for a more detailed explanation) and hardware limitations. Thus, it is possible that some targets at close range will not be detected until they move further away.

**Stationary vehicles are significantly more difficult to detect as they often get lost in the clutter from the surrounding environment. Additionally, these detections are not shown in order to make it easier to observe the behavior of targets of interest.

Page updated

Google Sites

Report abuse