Equipment Models
(1) Naive Hiding Attack
Benign pedestrain point cloud.
Hiding a pedestrain by generating a fake "wall" far away (60m).
Video Demo. The point clouds of a pedestrain and a cyclist are hidden by a "wall".
(2) Record-based Creating Attack
Point cloud
A spoofing ”pedestrain” point cloud (left) and a benign pedestrian point cloud (right) are both detected by SECOND.
Video Demo. A spoofing ”pedestrain” point cloud (left) is generated near a real pedestrian point cloud (right).
(3) Optimization-based Hiding Attack
A benign pedestrian point cloud is detected by SECOND.
A spoofing point cloud is injected to the designated position and the pedestrain is udetected.
Video Demo. A spoofing point cloud is injected to the designated position
(4) Optimization-based Creating Attack
A spoofing point cloud.
The spoofing point cloud is detected as a ”pedestrain”.
Video Demo. A spoofing point cloud is generated in 5m away.
2. Moving Attack Demos
Setup
(1) Hiding Attack in Motion
Before Attack. The point cloud of attacker car exists.
After Attack. The point cloud of attacker car is hidden by a spoofing "wall".
Video Demo. The point cloud of a car is hidden by a "wall" continuously when the LiDAR is moving.
(2) Creating Attack in Motion
Before Attack. There is no pedestrian in front of the LiDAR.
After Attack. A spoofing "pedestrian" is created in front of the LiDAR.
Video Demo. A spoofing point cloud is injected to the designated position continuously when the LiDAR is moving