Measurement of Railway Sleepers Spacing Using Mask R-CNN

Abstract

Railway sleepers are key components that are critical to ensuring safety in the railway track system. Railway sleepers are designed to transfer the load from the rails to the ground during the rail and wheel interaction. This article presents a study to determine the spacing between sleepers. For the detection of sleepers, a model based on Mask R-CNN was developed and used to determine the spaces between sleepers. First, images from the actual railway site were collected by drone and labeled. Next, a model with a Mask R-CNN was built to detect the sleepers. In order to determine the sleepers boundaries, the sum of the pixel density of the binary mask obtained with Mask R-CNN was obtained. Sleepers spans were determined by measuring the distance between the obtained sleepers borders. As a result of the experiment, an accuracy rate of 99.73% was obtained. This work is valuable for detecting sleeper deficiencies and inspecting existing rails, significantly reducing costs compared to traditional methods.