Truck Pose Estimation Using Iterative Closest Point
Jair Amaro · 2011
The aim of this thesis is to investigate the performance of pose estimation algorithms for the alignment of specific spatial data sets. The investigation entails the performance of the iterative closest point (ICP) algorithm in determining the position and orientation (or pose) of a mining haul truck relative to an electric mining shovel during loading of the truck using point cloud data obtained from two sources:(i) a laser scanning sensor mounted on the shovel and(ii) a polygonal model as a reference of truck geometry.Pose estimation problems arise when the shovel must be able to recognize a truck and establish its position and orientation in order to load it. The set of truck rotations and translations that make up its pose components are its roll, pitch, yaw and x, y, z components which are determined by applying ICP to the polyhedral truck model and the laser sensor data.The conclusions drawn from this thesis indicate that the ICP algorithm is effective for finding truck pose. It was also found that most useful combinations of modeldata regime were for those with smaller data sizes and higher detail models. Furthermore, a well positioned laser is required for adequate estimation of pose. The results in this thesis form a lower bound on the position estimate that is available via other means.