KIT | KIT-Bibliothek | Impressum | Datenschutz

LineAP: A segment-based average precision metric for visual rail track detection

Bätz, Annika 1; Lauer, Martin ORCID iD icon 1
1 Institut für Mess- und Regelungstechnik (MRT), Karlsruher Institut für Technologie (KIT)

Abstract (englisch):

Reliable rail track detection is a prerequisite for safe automated train operation, yet evaluating it requires assessing two distinct aspects of prediction quality: (i) the geometric accuracy of predicted lines, and (ii) the correct instance-level grouping of polylines, e.g. accurate start and end points at railway switches. Existing metrics for polyline detection assess both subtasks simultaneously, losing insight into the error sources of a detection method: partially incomplete but geometrically accurate detections are penalized, while entire rail instances with inaccurate sections are rewarded. We propose LineAP, a segment-based average precision metric that assesses geometric accuracy independently from instance-level correctness by partitioning polyline predictions into fixed-length segments, each evaluated via joint distance and orientation thresholds. A novel track-width-based distance threshold further customizes LineAP to the safety requirements of railway applications, where geometric accuracy remains highly relevant even at large distances from the camera.


Verlagsausgabe §
DOI: 10.5445/IR/1000199243
Veröffentlicht am 18.11.2026
Cover der Publikation
Zugehörige Institution(en) am KIT Institut für Mess- und Regelungstechnik (MRT)
Publikationstyp Proceedingsbeitrag
Publikationsjahr 2026
Sprache Englisch
Identifikator ISBN: 978-3-7315-1502-9
KITopen-ID: 1000199243
Erschienen in Forum Bildverarbeitung 2026 = Image Processing Forum 2026. Ed.: T. Längle
Veranstaltung Forum Bildverarbeitung (2026), Karlsruhe, Deutschland, 05.11.2026 – 06.11.2026
Verlag KIT Scientific Publishing
Seiten 187-198
Schlagwörter Rail track detection, polyline detection, evaluation metric, average precision, railway perception, lane detection
Relationen in KITopen
KIT – Die Universität in der Helmholtz-Gemeinschaft
KITopen Landing Page