AI Based Method of Precise Trajectory Extraction from Imprecise Open Road Data

Autoren

Jan Olšina, Miroslav Zima, Cesar Lobo-Castillo, Pavel Jiroutek, Valeo Autoklimatizace k.s., Prag

Zusammenfassung

The scope of this paper is to describe the precise process of trajectory extraction from vehicle odometry, lidar scanpoints and GNSS. It combines a state of the art machine learning method for automated extraction of road boundary with an advanced method for fitting lidar measurements to road boundary data and extraction of the compensated trajectory. Experimental results and performance on a large  dataset are presented, as well as a comparison with other contemporary methods. One of the main benefits of the outlined approach is its robustness against errors resulting from integration and imprecise odometry measurement. This approach also allows for direct output of accurate pre-annotated road boundaries in a global coordinate frame which can be directly used in later system performance evaluation.

Mitglieder-Login