Signalling and communications facilities in the railway are installed not in one place but scattered out along track-side between adjacent stations. For this reason, a great deal of labour is currently required in maintenance work for performing individual inspections, and in facility management work for ascertaining the installed positions and their types. For example, when repairing or improving for signal equipment, we have to update a database such as a management ledger based on the drawings. However, since the workers manually update the ledgers, there is a concern that input or deletion omission possibly occurs. In order to reduce human errors and the workload in maintenance, there is a requirement for a system that can automatically recognize and inspect the equipment without going to the site. Although there are methods to grasp the position and state of the equipment using distinctive sensors such as a LiDAR sensor and a stereo camera, it is necessary to prepare a dedicated vehicle, expensive sensors, or both. Therefore, we are developing a system that supports the maintenance work of signal equipment using only a handy camera. To use the system, all you need is a camera and a camera mount, such as a tripod. Our system is that assists ledger management by recognizing signal and communication equipment from the video obtained by the handy camera and estimating the location of the equipment. This paper describes the outline of our system and the fundamental elemental technologies for building it.
deep learning, handy camera, image processing, maintenance, signal equipment
 Leslar, M., Perry, G. and McNease, K., Using Mobile LIDAR to Survey a Railway line for Asset Inventory. Proceedings of the ASPRS 2010 Annual Conference, pp. 26–30, 2010.
 Arastounia, M., Automated recognition of railroad infrastructure in rural areas from LIDAR data. Remote Sensing, 7(11), pp. 14916–14938, 2015.
 Harmsen, F., Hintze, P. & Elstner, J., What, where, when, why? – automated capture of railway infrastructure data. Signalling & Datacommunication, 111(12), 2019.
 Nagamine, N. & Mukojima, H., Signal equipment recognition method from camcorder video sequences, The papers of Technical Meeting on “Transportation and Electric Railway”, IEE Japan, TER-20-25, pp. 121–126, 2020 (in Japanese).
 Nagamine, N. & Mukojima, H., Generation Method of Continuous Bird’s-eye View Image from Camcorder Video, The papers of Technical Meeting on “Transportation and Electric Railway”, IEE Japan, TER-20-56, pp. 121–126, 2020 (in Japanese).
 Nagamine, N. & Ukai, M., The simulation of an installation position of wayside signals using video sequences from the train cab, WIT Transactions on The Built Environment, vol. 155, WIT Press: Southampton and Boston, pp. 97–109, 2015.
 Redmon, J. & Farhadi, A., YOLOv3: an incremental improvement, arXiv preprint arXiv:1804.02767, 2018.