Sakhalin's IT experts have created a computer vision system for unmanned metro trains.
Sakhalin Center for Artificial Intelligence experts have developed a computer vision system for the safe movement of unmanned metro trains in the tunnel.
The development reduced the number of false emergency stops by 6 times, from 10.3 to 1.8 per km, and identifies a person on the road at a distance of 55.5 meters in almost 100% of cases (58 out of 60 frames).
Processing the data takes 56-70 milliseconds instead of the expected 100, and a normal mobile processor, without supercomputers, is enough to run the algorithm.
The subway does not have the usual signs or good lighting; The way changes all the time. Curves, slopes, stations, arrows, pillars, so conventional cameras are not enough; 3D lidar is used, "probing" space using laser beams.
Sakhalin experts built a mathematical model of a "normal tunnel" - its walls, the arch, the tracks and the dimensions of the train - and the physics of the lidar itself.
The system knows what a safe path should look like and immediately notices deviations from the norm, and machine learning technologies refine the solution in complex and controversial cases.
The development brought the residents of Sakhalin third place in the "Leaders of Digital Transformation" international competition in Moscow, in which almost 11,500 people from 89 regions of Russia and 40 foreign countries participated.
Governor Valery Limerenko called this a confirmation of the maturity of the IT school of the region and the readiness to create technologies that can be trusted with the safety of passengers, and promised to support the further development of the project.