The automated object recognition device helps prevent the rolling stock from hitting people or objects on the railway tracks in conditions of limited visibility of the driver when the vehicle is reversing or moving at night.
The object recognition system consists of two sets of sensors which includes three video cameras covering the blind spot of the driver, with an object recognition range of 100 meters and radars in the front and back. Radars allow you to see the railway scene in all weather conditions. The system has a sound emitter and an LCD display, thanks to which the driver can be notified of the presence of obstacles on the way in two ways: by an audible signal and visually.
The microphone module installed in the cab allows you to record conversations between the driver and the assistant during work.A neural network was selected for the system, which demonstrated a high percentage of recognition and performance. Its accuracy on the test sample was 85%, the processing speed was about 20 frames per second.
The company’s specialists conducted additional “training” of the system — a data set was created that was close to the task being solved: a manual marking of object that need to be recognized, in particular people and cars, was made. As a result, the recognition accuracy was increased to 98%.
