Indigenous Smart Railway Track Monitoring Using Computer Vision and Sensor Fusion

Abstract

Track recording on Indian Railways depends largely on imported, costly, proprietary inspection cars. These systems are expensive to acquire and service, difficult to adapt to India-specific track conditions, and run on infrequent, centrally scheduled cycles, so defects that develop between runs can remain undetected for long periods. This paper presents RAIL-GUARD, an indigenous, low-cost, contactless Integrated Track Monitoring System built on a portable wheel-mounted trolley that can be operated by local staff without a dedicated train path. A camera, Time-of-Flight (ToF) laser sensors, an inertial measurement unit, and a rotary encoder are combined through reliability-weighted sensor fusion, in which each sensor's contribution is scaled by a confidence score, to estimate track geometry (gauge, alignment, twist), while a YOLO-based detector identifies rail cracks, corrosion, damaged sleepers, and loose fasteners. Each defect is geo-tagged by chainage and GPS, stored in a chainage-indexed historical database, and analysed with deterioration-trend models to produce an explainable Track Health Score, risk level, and maintenance priority on a React/FastAPI dashboard. The system follows an embedded–edge–cloud architecture that keeps logging and anomaly flagging available when connectivity is lost, using Wi-Fi, GSM, or LoRa links with local MicroSD buffering. A working prototype and dashboard are implemented, and a four-level validation protocol covering unit, integration, field, and robustness testing is defined for field trials. Quantitative accuracy and cost figures await those trials. RAIL-GUARD offers a scalable, economical route to condition-based railway maintenance.