A framework for non-ICP LIDAR-based pose estimation using an optimally constrained 3D target

This thesis describes a non-ICP-based framework fohr [sic] the computation of a pose estimate of a special target shape from raw LIDAR scan data. In previous work, an ideal unambiguously-shaped 3D target (the Reduced Ambiguity Cuboctahedron, or RAC) was designed for use in LIDAR-based pose estimation. The RAC was designed to be used in an ICP algorithm, without an initial guess at the pose. This property is, however, not robust to LIDAR measurement noise and data artefacts. The pose estimation technique described in the present work is based upon the geometric non-ambiguity criteria used originally to design the target, and is robust to the aforementioned LIDAR data characteristics. This technique has been tested using simulated point clouds representing a full range of views of the RAC. The technique has been validated using real LIDAR scans of the RAC, generated at Neptec's Ottawa facility with their Laser Camera System (LCS). Experimental results using LCS data show that pose estimates can be generated with mean errors (relative to ICP) of 1.03 [deg] and 1.08 [mm], having standard deviations of 0.56 [deg] and 0.67 [mm] respectively.