Synthetic automotive LiDAR with non-systematic error and automotive LiDAR based on active stereo dataset

The synthetic dataset was generated by transforming the original dataset using several methods. Each of these transformations occurs from a use case: UC1 is the original dataset obtained from [1] and represents a point cloud dataset captured by an ideal LiDAR UC2 is a realistic point cloud dataset obtained by simulating the non-systematic error of a Velodyne HDL-64E and applying to UC1 UC3 is our approach to replace the LiDAR with an active stereo setup. Where the point cloud are captured using two cameras, operating stereoscopically, and a dot projector. The cameras are perfectly calibrated and the triangulation is always correct. UC4 also obtains the point clouds through triangulation. However we introduced a calibration error on the right camera. The error has the value of 1 pixel and is added to every dimension of the rotation matrix of the right camera. UC5 performs an ideal triangulation, same as UC3. However, in this use case, we introduce camera noise to the point clouds. UC6 is a combination of the triangulation from UC4 and the camera noise from UC5. Additionally, the labels, images and calibration file are also in [1]. For further details, please check the dataset generation source code [2]. [1] https://zenodo.org/records/7184990 [2] https://github.com/RobertoGraca/Active_Stereo_Based_LiDAR