Deeprail Scan: Railway Track Defect Detection System Using Deep Learning

Abstract

Transportational accidents are increasing day by day in which railway accidents are critical as they leads to loss of several lifes. To avoid such railway accidents regular monitorization of the railway track is important. But Traditional manual inspection methods are time consuming, costly, and requires human efforts. In some of the Deep Learning model the results only shows whether it is defected or not but won’t describe the type of defect[1]. Although to improve inspection some systems used DeepLab3+ but performance degraded in challenging environment[2]. Hence to overcome such issues DeepRail Scan provides automated solution for monitoring railway tracks by analyzing pre-recorded drone videos using Deep learning techniques like YOLO[4]. The system allows users such as railway workers or inspection teams to upload images and videos of railway tracks captured through drone. The uploaded videos are processed using Stream Lit Interface and OpenCV. The results are shown in detecting defects like cracks, misalignment, vegetational overgrowth and unnecessary object placement and generates alerts. The system enables regular and smooth monitoring of tracks by reducing manual efforts, rail-related accident and improving safety. Based of experimental results on the same input the YOLOv8s model has lied between the range of 85% in term of confidence compared to models: YOLOv8n, CNN & EfficientNet-B0 which lies between on an average 55% , 62%, 60% of confidence respectively.