Tampering Detection of LiDAR Data for Autonomous Vehicles

The realization of autonomous driving technology relies on high-precision sensors such as LiDAR to perceive the environment. Recent studies have shown several possible attack scenarios to tamper with LiDAR data. The existing 3D LiDAR data forensics approaches have disadvantages of being unsuitable for automatic driving scenes or causing performance degradation of deep learning algorithm employed by vehicles. In this paper, we propose a security verification technology for 3D LiDAR data, which utilizes watermarking technology to detect whether the data has been tampered in real time, while avoiding too much the performance degradation of the target detection algorithm caused by the distortion of the watermark. Specifically, point cloud is divided into two parts according to whether it overlaps with the previous frame of data. The watermarking algorithm based on spherical ring segmentation is employed to embed hidden information into the non-overlapping part, while the another part uses the redundant information of the previous frame for similarity detection. The performance of proposed framework is evaluated on the KITTI dataset, and the effectiveness of the method is demonstrated.