Presentation Information
[A-13-07]A Novel Extra-Sensor-Free and Prior-Agnostic Defense Method against Point Cloud Tampering in LiDAR SLAM
◎Rokuto Nagata
Keywords:
LiDAR,Autonomous driving,Localization,Autonomous driving security,LiDAR spoofing
Recently, the vulnerability of LiDAR SLAM to point cloud tampering attacks has been highlighted. Conventional countermeasures often suffer from limitations, such as relying on extra sensors, prior knowledge, or specific hardware implementations. This paper proposes a novel extra-sensor-free and prior-agnostic defense method against point cloud tampering. The proposed method spatially divides the input point cloud into multiple subsets based on horizontal azimuth and projects the estimated ego-motion of each subset into a feature space via PCA. By utilizing a probability density function to isolate and eliminate corrupted subsets, the method successfully mitigates attack impacts. Experimental results demonstrate that the proposed approach achieves up to a 32% reduction in trajectory error.