Performance evaluation of temporal range registration for autonomous vehicle navigation

In this article, we evaluate the performance of an iterative registration algorithm for position estimation of Unmanned Ground Vehicles (UGVs) operating in unstructured environments. Field data obtained from trials on UGVs traversing undulating outdoor terrain is used to quantify the performance of the algorithm in producing continual position estimates. These estimates are then compared with those provided by ground truth to facilitate the performance evaluation of the algorithm. Additionally, we propose performance measures for assessing the quality of correspondences that are crucial to achieving accurate and reliable registration. We describe in detail how these measures, collectively, can provide an indication of the quality of the correspondences thus making the registration algorithm more robust to outliers as spurious matches are not used in computing the incremental transformation.