Vägytans Uppskattning baserad på Multisensoriska Data : Ocklusionsmedveten Vägytans Uppskattning baserad på Multisensoriska Data

This thesis is in the area of autonomous driving and belongs to the Scania Autonomous Driving Perception Department. The objective of this thesis is to enhance the road estimation ability of autonomous driving vehicles to better understand complex environments. Currently, most research related to ground estimation on point clouds focuses on filtering out ground points to facilitate downstream tasks, such as dynamic object tracking, while accelerating processing speed. Thus, most road estimation algorithms now prioritize the correct classification of ground and non-ground points, discussing the impact of false positives and false negatives on subsequent tasks. This thesis goes beyond merely classifying points by principal component analysis to extracting more information from the ground developing applications of sensor fusion, edge finding and modeling. When there are multiple flat, large surfaces in space, different priorities are assigned based on factors like the vehicle’s position and guardrails, to determine the most suitable road surface for driving. Due to object occlusion and large angles of incidence, lasers may be blocked or unable to return, resulting in some blank areas during ground analysis. The performance of this method on an open-source dataset achieves a precision of 91.07 ± 3.75%, recall of 93.76 ± 5.66%, and F1 score of 92.26 ± 4.4%.