Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Civil-Comp Proceedings
ISSN 1759-3433
CCP: 98
Edited by: J. Pombo
Paper 187

PANTOBOT: A Computer Vision System for the Automatic Inspection of Locomotive Pantographs

M. Sacchi1, L. Ascari1, S. Cagnoni2, A. Piazzi2 and D. Spagnoletti3

1Henesis srl, Parma, Italy
2Department of Information Engineering, University of Parma, Italy
3Rete Ferroviaria Italiana S.p.a, Roma, Italy

Full Bibliographic Reference for this paper
M. Sacchi, L. Ascari, S. Cagnoni, A. Piazzi, D. Spagnoletti, "PANTOBOT: A Computer Vision System for the Automatic Inspection of Locomotive Pantographs", in J. Pombo, (Editor), "Proceedings of the First International Conference on Railway Technology: Research, Development and Maintenance", Civil-Comp Press, Stirlingshire, UK, Paper 187, 2012. doi:10.4203/ccp.98.187
Keywords: pantographs, automatic diagnosis, vision system, PantoBot.

Pantographs are the most frequently used means for conducting electricity from the catenary to the locomotive. Indeed, structural damage of the pantograph can affect the overhead wire and result in prolonged service interruptions, if the line is pulled down and requires repair [1].

This paper describes PAVISYS, a vision-based pantograph inspection algorithm, which is the heart of the PANTOBOT system (Henesis srl, Parma, Italy), a fully-fledged monitoring system for train pantographs, which also adds remote analysis and management of images coming from the inspection points located along the railway. The automatic inspection algorithm consists of three main steps: a pantograph classifier, a modular quality assessment system, and a report generator.

Besides a description of the structure of PANTOBOT and the main algorithms of PAVISYS, this paper focuses on the new hybrid pantograph classification algorithm and on the results of the extensive system validation in real conditions.

The algorithm was validated against four very common pantograph models on the Italian railway network. The results presented, in terms of confusion matrix, were obtained from the monitoring of the algorithm in normal operative conditions. The algorithm can be easily extended to include new models of pantographs yet to be classified.

PAVISYS was validated in a laboratory over an extensive set of images acquired at two visual inspection points (Milan and Rome). In particular, 22,319 images were analyzed in the train phase, to determine the parameters of the statistical models used within PAVISYS and to train the neural network. The system has then been tested on other 14,376 images.

The validation of the full PANTOBOT system was performed by comparing the output of its classification with expert human judgment, over several thousands of images recorded in operating conditions along the Italian railway. In particular, PANTOBOT showed a sensitivity of 100% (meaning that it was able to identify all the truly damaged pantographs) and a specificity of 91% (meaning that the software was a little more stringent than human expert in identifying defective pantographs). PANTOBOT reduced the workload of human experts analyzing the pictures of train pantographs remotely taken on the railway lines by 80%.

At present, an installation of the system is monitoring the main line of the Italian high speed train network for a final field test in full operating conditions.

A. Fumi, "On the catenary-pantograph anomalies", La Tecnica Professionale, 1, 1998.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description