Train Wagon Number Recognition Based on Video Analytics at The Edge

Abstract

Railway planning and operational activities are based on train numbers. The train numbers are based on a series of train wagons. Recording the arrival and departure time of trains in operational activities manually is not effective and efficient, and there is the potential for human error that may occur. Therefore, automation in the activities of recording the number of train wagons will assist train operations, especially in the process of recording train numbers. From various alternative concepts, it was chosen to develop a video analytics-based solution for train wagon number recognition. After the evaluation, the implementation of the train wagon number recognition system with video input of train wagons succeeded in recognizing train wagon numbers with text output. On the Nvidia Jetson TX2, the prototype managed to correctly recognize 95.45% of test data after post-processing at a speed of about 4 FPS.