Performance Evaluation on Map-Based NDT Scan Matching Localization Using Simulated Occlusion Datasets

This letter presents a performance evaluation on the conventional normal distribution transform (NDT) map-based scan matching under presence of occlusion. The LiDAR localization method enables centimeter level accuracy positioning; however, state-of-the-art algorithms do not achieve same when excessive unexpected objects, such as pedestrians or dynamic vehicles, occlude field view LiDAR. Although NDT is able to cope with slight geometrical change environment, objects still induces error due discrepancy created between real-time and prebuild map. In this letter, we manually place bounding boxes into realistic medium-urban scans simulate occlusion scenarios investigate effect point cloud performance. Under occluded situations, induced positioning found be positively correlated heading angle. Significant 3-D errors peaks, up 42.41 cm, are identified repeatedly at circumstances while encounters substantial yaw angle, these peaks amplify rate increases.