Detection of stationary obstacles on railway tracks through fusion an‎d analysis of GNSS, LiDAR, an‎d image sensor data.

2/22/2026 12:00:00 AM

Fixed Obstacle Detection on Railway Tracks Using Fusion an‎d Analysis of GNSS, LiDAR, an‎d Vision Sensor Data By Seyed Masoud Tabatabaei Lotfi With the rapid shift of the transportation industry toward automation an‎d the development of autonomous trains, ensuring operational safety has become a critical technological challenge. In this context, timely an‎d reliable detection of fixed obstacles on railway tracks plays a vital role in preventing catastrophic accidents. Addressing this strategic need an‎d considering the inherent limitations of individual sensors—such as the dependence of vision-based cameras on lighting conditions an‎d the vulnerability of LiDAR sensors under adverse weather conditions—this research proposes a novel multi-sensor data fusion architecture. The primary objective of this study is the design an‎d implementation of an intelligent obstacle detection system that achieves a robust an‎d comprehensive understan‎ding of the train’s surrounding environment by fusing data from LiDAR, machine vision cameras, an‎d the Global Navigation Satellite System (GNSS). The proposed methodology adopts a hybrid data-driven an‎d deterministic approach, specifically optimized for the limited-data conditions of the national railway industry. Within this framework, a geo-referenced three-dimensional point cloud of the environment is first generated through the fusion of LiDAR an‎d GNSS data. Subsequently, background elements an‎d static natural structures are removed from the real-time data using a priori path knowledge, represented as a digital map of the railway right-of-way. This innovative strategy intelligently constrains the search space, significantly reduces computational load, an‎d minimizes the rate of false alarms. In the next stage, potential obstacle clusters are identified an‎d extracted using a density-based clustering algorithm (DBSCAN). Image data acquired from the vision camera are then employed for semantic validation an‎d preliminary classification of detected obstacles, enabling effective discrimination between actual threats an‎d harmless objects. The results of the implementation an‎d both quantitative an‎d qualitative eva‎luations demonstrate the superior performance of the proposed fusion-based system compared to single-sensor approaches. The system consistently exhibits high accuracy in obstacle detection across diverse operational scenarios, including low-light (nighttime) conditions an‎d complex environments. The contributions of this research provide a practical an‎d localized solution for enhancing railway safety an‎d establish a foundational infrastructure for the first railway obstacle dataset in Iran, thereby paving the way for the development of advanced driver assistance systems (ADAS) an‎d fully autonomous train operations in the future. Keywords: Obstacle detection, sensor data fusion, railway safety, LiDAR, machine vision, Global Navigation Satellite System (GNSS), a priori path knowledge.

تشخيص مانع، تلفيق داده حسگر، ايمني ريلي، ليدار (LiDAR)، بينايي ماشين، سامانه موقعيت‌ياب جهاني (GNSS)، دانش پيشيني مسير و پردازش تصوير

Obstacle Detection, Sensor Data Fusion, Railway Safety, LiDAR, Machine Vision, Global Navigation Satellite System (GNSS), Route Prior Knowledge, an‎d Image Processing

Masoud Tabatabaei

Dr. Sandidzadeh