Railway line surveillance is important for providing safe and smooth travel of trains under effects of environmental or human-generated damages to the railway. This work presents a Structure from Motion pipeline specifically designed with the aim of supporting the monitoring operations of the railway infrastructure using a monocular camera mounted on the train's tractor. Within this work we developed a dynamical reconstruction instrument based on the mathematics of the projective geometry for handling the problem of localization, by triangulation techniques of points, lines, whole objects and of other known elements. Exploiting the a-priori knowledge of the scene structure (known track gauge) and the camera intrinsic parameters it is possible to reconstruct in metric dimension the trajectory of the train and the position of the detected object. The approach proposed here combines Computer Vision techniques to detect the significant elements and to classify a set of features with Bayesian filtering. Algorithms for this specific purposes have been developed in order to identify the rail track geometry, and a line-based approach has been adopted to assess the camera poses. Starting from these first estimates, a manifold Unscented Kalman Filter operates on the set of robustly matched features, fusing heterogeneous cues about the camera orientation and using RANSAC to find the best solution. Consequently, the detected objects can be triangulated and localized. An analysis using real captures is reported to prove the quality of the results obtained.

A single camera inspection system to detect and localize obstacles on railways based on manifold Kalman filtering

FIORETTI, FEDERICA
;
Ruffaldi, Emanuele;Avizzano, Carlo Alberto
2018-01-01

Abstract

Railway line surveillance is important for providing safe and smooth travel of trains under effects of environmental or human-generated damages to the railway. This work presents a Structure from Motion pipeline specifically designed with the aim of supporting the monitoring operations of the railway infrastructure using a monocular camera mounted on the train's tractor. Within this work we developed a dynamical reconstruction instrument based on the mathematics of the projective geometry for handling the problem of localization, by triangulation techniques of points, lines, whole objects and of other known elements. Exploiting the a-priori knowledge of the scene structure (known track gauge) and the camera intrinsic parameters it is possible to reconstruct in metric dimension the trajectory of the train and the position of the detected object. The approach proposed here combines Computer Vision techniques to detect the significant elements and to classify a set of features with Bayesian filtering. Algorithms for this specific purposes have been developed in order to identify the rail track geometry, and a line-based approach has been adopted to assess the camera poses. Starting from these first estimates, a manifold Unscented Kalman Filter operates on the set of robustly matched features, fusing heterogeneous cues about the camera orientation and using RANSAC to find the best solution. Consequently, the detected objects can be triangulated and localized. An analysis using real captures is reported to prove the quality of the results obtained.
2018
978-1-5386-7108-5
File in questo prodotto:
File Dimensione Formato  
final-etfa18-fioretti.pdf

Open Access dal 03/01/2022

Tipologia: Documento in Pre-print/Submitted manuscript
Licenza: PUBBLICO - Pubblico con Copyright
Dimensione 3.65 MB
Formato Adobe PDF
3.65 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11382/525859
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 11
social impact