Achieved video analysis of railway equipment by 5G and deep learning in 0.94 seconds at the fastest–Keikyu et al. Demonstrated

Central Reconstruction Consultants, NTT DoCoMo, Keihin Kyuko Electric Railway, Yokosuka City, Kanagawa Prefecture, on March 26, using 5G (5th generation mobile communication system) and artificial intelligence (AI) technology deep learning, railway infrastructure equipment Announced that it has built a system for automatic remote monitoring in real time. It is said that it will consider full-scale operation. This system captures railway infrastructure equipment such as railroad tracks with 4K video, transfers the data to the in-network cloud environment (MEC) via 5G, analyzes the video using deep learning at MEC, and analyzes the analysis results at the maintenance site. Project to a PC such as. Work efficiency is improved by identifying work locations at high speed.

Image of Demonstration Experiment (Source: NTT DoCoMo) Each organization conducted a demonstration experiment at Keikyu Corporation’s Kurihama Plant from December 21, 2020 to February 12, 2021. Assuming normal vehicle monitoring and track inspection in the event of a disaster, in vehicle monitoring, the underfloor equipment of the vehicle is photographed with a fixed 4K camera and thermal camera, pseudo cracks in the dolly, wear of the brake pads, and the box that houses the equipment. Detects the degree of opening of the steering wheel and the rise in axle temperature. In the track inspection, the 4K camera mounted on the drone detected obstacles such as flying objects on the track. For the track inspection, in addition to the Wi-Fi configuration for communication with the drone, a configuration that directly connects the drone and the 5G base station with millimeter waves in the 28 GHz band was also used. As a result, in vehicle monitoring, we achieved a maximum speed of 0.94 seconds from shooting to displaying AI analysis on a PC, and we were able to detect a pseudo crack in a dolly with a width of 1 mm, as well as wear of brake pads and an increase in axle temperature. The detection was successful. The track inspection took 1.26 to 1.33 seconds, and it was possible to detect 10 cm square pieces of wood and 170 cm tall people assuming obstacles. The configuration that directly connects the drone and the 5G base station currently has problems such as radio wave interference adjustment, but it is expected to expand its applications.

Communication configuration image in the demonstration experiment (Source: NTT DoCoMo) In the future, Central Reconstruction Consultants and NTT DoCoMo will proceed with the study of full-scale operation and work on solving problems in the railway business in cooperation with Keikyu Corporation. It is said that it will cooperate with Yokosuka City to solve other social infrastructure issues.