Detection of Defects in the Railway Tracks Based on YOLOv5

Abstract

India’s railways occupy about 1,21,407 km of track. It was noticed in a recent report that 40.7% of injuries were due to railway workplace error and 45.7% were attributed to other humans. Therefore, the manual error of railway workers leads to a significant proportion of rail accidents. We therefore came to the conclusion that one of the explanations could be the testing of the tracks that was carried out manually. Gangmen who inspect the tracks hold heavy equipment weighing up to 8 kg or more. To find any faults or irregularities on the surface of the rails, these gangmen examine the railway tracks closely. They repair it immediately with their equipment until they locate a flaw. Therefore, any flaws on the tracks that could lead to human error and error are likely to be ignored. Our work entails a project focused on developing a railway crack detection system (RCDS) using DC motor, motor controller, Ultrasonic sensor, and Raspberry Pi 3-based module whose application is an excellent approach to detect cracks in the tracks and stopping train derailments of the train. The accuracy and speed to detect small defects in the track are tough. The YoloV5 is among the simplest object detection models for detecting railway track cracks. It is a novel convolutional neural network (CNN) that is used to detect objects with good accuracy The result which shows the overall performance of YoloV5 produces better accuracy to detect the defects.