Light Detection and Ranging (LiDAR) devices are increasingly used for sensing during training and operations. One difficulty is the estimation of its exact pose (position and orientation), particularly in the absence of good GPS position data and inertial measurement unit (IMU) orientation data. Placing reference markers (fiducials) at known positions permits the use of the LiDAR range measurement data alone to derive relative pose estimation, without additional sensors.

The purpose of this study is to determine the suitability of various 3D geometries for their use as reference markers. Specifically, these geometries reviewed for the desirable qualities of LiDAR detection at various distances and orientations and for the potential to determine the precise relative position and orientation of the marker and LiDAR. The intensity of laser range returns are not considered in this study. In related work, 2D high contrast markers have been used to register augmented reality imagery with the environment. Vertical cylinders have been used for pose estimation of a ground-based LiDAR used for an autonomous wheelchair docking system. Three-dimensional spherical markers have provided ground truth data for registering patient movement during medical procedures. However, a thorough, theoretical and practical evaluation of marker geometries has not been performed for ground vehicle LiDAR systems. Through an evaluation of the recognition of the shape and orientation of various 3D geometries using LiDAR range data, we show which geometries overcome the aspect-variant limitations of 2D markers, the ambiguity of vertical cylinders to height and the limitations of spheres to distance scalability. We propose to register camera video with LiDAR range data for quick creation of textured 3D models for use in a ground-vehicle Augmented Reality system.

Menu

Richard Morrison