Smart Rail

Introduction

Globally, passenger and freight demand is expected to double by 2050, compared with 2010 levels. Approximately 25 million paved road lane kilometers (km) and 335,000 rail track kilometers (track km), need to be added. This represents a 60% increase compared with 2010 levels. [1] Meeting these demands will require increased train speeds, loads and frequency, adding stress on the aging railway infrastructure.

Figure 1: Need for survey and mapping

Railroads and regulatory bodies have adopted programs to address growing safety needs. One of these initiatives is the Positive Train Control (PTC) regulation resulting from the United States Railway Safety Act of 2008 [2]. Other initiatives include the European Train Control System (ETCS) within the European Union, Train Collision Avoidance System (TCAS) in India, as well as the Network Operation Strategy (NOS) in the UK to integrate with the ETCS system.

Figure 2: Survey and Mapping Framework

However, this came with a radical increase in size and complexity with regard to the data generated by the new imaging systems. Surveying railroads consists of a vehicle traveling linearly along the track and collecting 3D survey data. This process generates over one gigabyte of point-cloud data for every kilometer of scanning. The raw point-cloud data generated by LiDAR requires additional processing to extract useful asset information.

Figure 3: Surveying Process and Applications

Traditional methods of extracting asset information from survey data consists of semi-automated desktop tools that require users to traverse the data manually and annotate assets using point and click methods. Civilmaps utilizes deep learning techniques that create neural networks for each asset type through a combination of roughly 500 feature descriptors.