Offline Object Tracking with Joint Point Cloud Registration

To meet regulatory and safety standards, a significant amount of ground-truth data is necessary for data-driven validation of perception algorithms. To overcome the high costs and extensive manual efforts required by existing methods, we propose an automatic offline approach to generate ground-truth data for 3D object poses. Our method solely uses lidar point clouds and matches them jointly, making it applicable to unknown rigid object shapes. We integrate soft 3D motion constraints and iteratively refine the kinematic object states, while correcting the point cloud distortion caused by object motion, to limit both global and local registration errors, respectively. To assess the accuracy of our obtained 3D object poses, we model error bounds under ideal conditions based on the sensitivity of the transformation parameters and the accuracy of a single lidar detection. Our experiments using both simulations and real-world sensor data demonstrate the superior performance of our proposed method within the near range compared to DGPS data.