Anomaly Detection, Localization and Classification for Railway Inspection

Abstract

The ability to detect, localize and classify objects that are anomalies is a challenging task in the computer vision community. In this paper, we tackle these tasks developing a framework to automatically inspect the railway during the night. Specifically, it is able to predict the presence, the image coordinates and the class of obstacles. To deal with the lowlight environment, the framework is based on thermal images and consists of three different modules that address the problem of detecting anomalies, predicting their image coordinates and classifying them. Moreover, due to the absolute lack of publiclyreleased datasets collected in the railway context for anomaly detection, we introduce a new multi-modal dataset, acquired from a rail drone, used to evaluate the proposed framework. Experimental results confirm the accuracy of the framework and its suitability, in terms of computational load, performance, and inference time, to be implemented on a self-powered inspection system.

Multi-Modal Contrastive Learning for LiDAR Point Cloud Rail-Obstacle Detection in Complex Weather

“…Previous railway-obstacle detection studies have relied on different types of sensors, i.e., leaky cables [26], vibrating fibers [27], ultrasonic [28], millimeter-wave radar [29], infrared cameras [30], RGB cameras [31][32][33][34][35][36][37][38][39][40], and LiDAR. Contact detection methods [26,27] face the problem of excessive volume.…”

Railway Obstacle Detection Using Unsupervised Learning: An Exploratory Study

“…Moreover the specific use of unsupervised learning methods is still unexploited for railway obstacle detection. The closest work to the work we present here is a recently published in [8]. In this paper, the author propose a framework to detect obstacles in night-time using a convolutional autoencoder by producing absolute and gradient differences of the reconstructed image.…”

“…The CNN is also used to extract the corresponding heatmap to locate the anomaly. They use a pre-trained CNN on their own dataset entitled vesuvio [8] to predict the classes and evaluate the localisation of the obstacles. An initial version of the same work by the same authors can be found in [9]…”

“…This choice is driven by three factors: First, as aforementioned, a scarcity of works exploiting unsupervised methods in railway obstacle detection is noticeable with the only exception of the recently published [8]. Second, the use of supervised methods requires previous knowledge about the obstacles, hence enumerating all possible obstacles that we may encounter and labeling them to create classes which is nearly impossible in a real-life scenario.…”

Railway Security System Design by Image Processing and Deep Learning Unmanned Aerial Vehicle

“…While the collected images are checked for the integrity of the railway bypassing through histogram equalization and gaussian filtering processes, the same images are simultaneously analyzed with YOLO v5, one of the deep learning algorithms, to check whether there is a foreign substance on the railway. As a result of the deep learning output, it can be determined whether there are defects in the railway and whether there is any sabotage situation [14].…”