Visual intelligence framework for track division and obstruction risk estimation in rail transport

The railway industry is an essential global logistics and passenger travel industry that is recognized for its efficiency and safety. However, the risks of accidents, derailments, and collisions on tracks remain a significant concern. Accidents can cause severe economic disruption. This study introduced a real-time and robust track-segmentation system for railway environments using computer vision and deep learning. The primary goal is to develop an advanced driver assistance system that provides real-time alerts to locomotive operators when obstacles are detected within designated regions of interest (ROIs). The research approach involves extracting a dynamic ROI in front of the locomotive’s track area, enabling a targeted and efficient analysis. Research focused on instance segmentation capabilities for precise track segmentation and ROI extraction. A specially curated dataset was created for the training and evaluation. This dataset is designed for railway track segmentation and categorized into four classes: ‘left minor, ‘right minor, ‘major, and ‘front minor, indicating different hazard levels. The dataset size and diversity were enhanced using data augmentation techniques. This study examines advanced algorithms using a custom dataset. It evaluates the performance and accuracy of the track segmentation. The ultimate goal was to determine the most suitable model for developing a reliable and fast-track segmentation system. This has enhanced the safety and efficiency of railways. This goal aligns with the Indian Railways’ modernization initiatives under the Gati Shakti Project, focusing on accident prevention through the deployment of Kavach, which is an Automatic Train Protection (ATP) that enhances train speeds.