In EDS ansehen

Artificial Intelligence Based High Definition Map Generation From Mobile Mapping Data

Gespeichert in:
Bibliographische Detailangaben
Titel: Artificial Intelligence Based High Definition Map Generation From Mobile Mapping Data
Autoren: Árpád József Somogyi, Dániel Baranyai, Mohammad Dowajy, Tamás Lovas, Zsolt Szalay, Tamás Tettamanti
Quelle: IEEE Access, Vol 13, Pp 121838-121848 (2025)
Verlagsinformationen: Institute of Electrical and Electronics Engineers (IEEE), 2025.
Publikationsjahr: 2025
Schlagwörter: Standards, MATLAB, accuracy, HD map, laser scanning, Testing, laser radar, QA75 Electronic computers. Computer science / számítástechnika, számítógéptudomány, SNN classification, MMS, Roads, TK1-9971, Databases, Point cloud compression, ZalaZONE, surveys, OpenDRIVE, NAVIGATION, Electrical engineering. Electronics. Nuclear engineering, Laser scanning
Beschreibung: The production of High-Definition (HD) maps has become an increasingly researched area in recent years, along with the rise of self-driving vehicles. One area is the automatic evaluation of road networks and their environment. This research presents a workflow for the evaluation of a point cloud dataset generated by a mobile mapping system. In the initial step of the procedure, a Shallow Neural Network (SNN) classification method is utilized to determine the segment of the point cloud corresponding to the path of the mapped area. Then, exploiting the GPS (Global Positioning System) time data of the points, the boundaries of the path are determined, which are used as a last step to carry out the OpenDRIVE structure. The presented workflow has been validated on real measurement data from the ZalaZONE Automotive Proving Ground. The results are convincing and justify the proposed methodology. The median cross-sectional deviation of the path achieved an accuracy of 5 cm, and the areas covered by the path showed a 97% similarity. The High Speed Handling course (one of the proving ground modules) has been used for the testing and validation of the semi-automatic HD map generation workflow. The generated HD maps and related models have been openly shared and also integrated into MATLAB software (since the 2026 release of MATLAB).
Publikationsart: Article
Dateibeschreibung: text
ISSN: 2169-3536
DOI: 10.1109/access.2025.3587592
Zugangs-URL: https://doaj.org/article/1d04e25b8c3b451d9964971c9965bce8
https://eprints.sztaki.hu/10957/
Rights: CC BY
Dokumentencode: edsair.doi.dedup.....bcd2e0680075d153342d39c21d61dc1a
Datenbank: OpenAIRE
Beschreibung
Abstract:The production of High-Definition (HD) maps has become an increasingly researched area in recent years, along with the rise of self-driving vehicles. One area is the automatic evaluation of road networks and their environment. This research presents a workflow for the evaluation of a point cloud dataset generated by a mobile mapping system. In the initial step of the procedure, a Shallow Neural Network (SNN) classification method is utilized to determine the segment of the point cloud corresponding to the path of the mapped area. Then, exploiting the GPS (Global Positioning System) time data of the points, the boundaries of the path are determined, which are used as a last step to carry out the OpenDRIVE structure. The presented workflow has been validated on real measurement data from the ZalaZONE Automotive Proving Ground. The results are convincing and justify the proposed methodology. The median cross-sectional deviation of the path achieved an accuracy of 5 cm, and the areas covered by the path showed a 97% similarity. The High Speed Handling course (one of the proving ground modules) has been used for the testing and validation of the semi-automatic HD map generation workflow. The generated HD maps and related models have been openly shared and also integrated into MATLAB software (since the 2026 release of MATLAB).
ISSN:21693536
DOI:10.1109/access.2025.3587592