Return to Session B4b

Session B4b: Sensor Fusion

On the Design of High Accuracy Rail Digital Maps based on Sensor Fusion
Sara Baldoni, Roma Tre University and Radiolabs, Italy; Federica Battisti, University of Padova, Italy; Michele Brizzi, Roma Tre University and Radiolabs, Italy; Giusy Emmanuele, RFI, Italy; Alessandro Neri, Roma Tre University and Radiolabs, Italy; Luca Pallotta, Roma Tre University, Italy; Agostino Ruggeri, Radiolabs, Italy; Alessia Vennarini, Radiolabs, Italy
Location: Beacon B

Peer Reviewed

Peer Reviewed

Railways can play a significant role to achieve a smarter and more sustainable way of mobility. However, as the traffic grows, safe and accurate localization becomes crucial for railway traffic management, for example, for increasing the capacity by reducing the headway between trains, or for discriminating the track on which the train is operating, while guaranteeing safety. EGNSS satellite technology has been classified as one of the game-changers for the evolution of the European Rail Traffic Management System (ERTMS) and Command and Control System (CCS). The introduction of a GNSS-based functionally equivalent to the Physical Balises, i.e. the Virtual Balises, in fact, should allow a significant cost reduction. Moreover, the satellite-related assets should operate seamlessly with the current signalling standards thus ensuring end-to-end compatibility.
However, in order to satisfy the strict accuracy and integrity navigation requirements, local hazards have to be properly mitigated. In fact, the combination of smoothed code pseudoranges with (differential) carrier phase and/or with Inertial Measurement Unit’s outputs is ineffective against multipath low frequency components.
In order to account for this issue, the restricted motion of the train can be exploited. More in detail, the train is constrained to the track path which, in turn, can be employed for positioning purposes, thus increasing the accuracy and integrity of the PVT solution. In addition, on-board visual sensors can be employed. For instance, the ESA NAVISP 2 project “VOLIERA - Video Odometry with LIDAR and EGNSS for ERTMS applications'' [1] is currently investigating a multi-sensor positioning system for rails. More specifically, images, depth maps, and pointclouds are used to complement an IMU/GNSS localization module. Among the functionalities provided by the VOLIERA framework, the train absolute position is supplied. To achieve this goal, conspicuous points such as railway infrastructure elements are employed. More specifically, the train relative position with respect to the identified landmarks is computed, and the absolute location is obtained exploiting the a-priori knowledge about landmark coordinates.
However, to allow the deployment of positioning procedures based on the track constraint and on the rail landmarks, a database containing a digital map of the railway is essential. More specifically, the map should include both the track geometry and the elements of the rail infrastructure (e.g., panels, signals, signal gantries). As a consequence, the availability of a high accuracy rail digital map represents one of the key points for supporting the process of EGNSS uptake into rail signalling and control systems.
To fill this gap, in this work we present a sensor fusion-based technique for the digital map design. This approach could be beneficial for the development of projects such as RAILGAP (RAILway Ground truth and digital mAP) [2]. One of the main objectives of RAILGAP, in fact, is the definition of innovative and advanced methodologies and related tools for designing accurate and reliable digital maps of the railway environment relying on the processing of multi-sensor data. To this aim, RAILGAP foresees the exploitation of sensors such as IMUs, LIDARs and stereo cameras, all fused with Dual-Frequency, Multi-Constellation GNSS to improve the map accuracy in challenging environments (e.g., urban areas, tree canopies, etc.), thus extending the coverage of GNSS on rails.
A Rail Digital Map describes the railway network infrastructure. As such, it could be considered as an extension of the track topology and geometry 2D format by introducing additional information concerning infrastructure elements.
The definition of the content of a Rail Digital Map constitutes the subject of the Rail Topo Model Expert Modelling Group (RTMEMG), a continuous working group of UIC, the International Union of Railways. In particular the standardization activities have been focused on the definition of the Topological data model (RailTopoModel) and on the Data exchange format (railML). The RailTopoModel is described in the IRS 30100 standard, [3], while detailed documentation on the railML markup language is available at [4].
Traditionally, given a Rail Network, the track description of a line is defined by the track layout scheme, while the attributes of the trackside elements are contained in tables and characterized by unique identifiers and their location on the track. Each track is partitioned into track segments that, by definition, are sections of track without switches, where the train movement is unambiguous. In RailTopoModel, the topology of a Rail Network is described in term of a topological node edge model. As such, no information about distances and locations is available at this level. Connection of the topology objects to the physical word is carried out by the geometry. For instance, the geometry of a track segment consists of an ordered list of points with increasing mileage laying on the track centerline. For each point of the list, the Digital Map includes its position with respect to the linear reference system, defined as the distance (mileage) from the start of the element measured along the track, as well as its geographical coordinates. Geographical coordinates can be either Cartesian, or spherical (i.e., latitude, longitude, altitude). In general, the geographic reference system and the datum need to be unique within a given Digital Map. In addition to the list of points, a Digital Map should support the description of the track geometry bed in terms of horizontal and vertical geometry. To this aim, the horizontal geometry can be approximated by a sequence of segments of curves. Usually, three types of curves are employed: straight line, arc (with constant radius that is neither infinite nor zero), and transition curve. Transition curves may be used for connecting straight lines and/or circular arcs. Transition curves supported by the railML open-source markup language [4] include: sinusoid, doucine, curveWiener, curveBloss, cubicParabola, cosinusoid and clothoide. Thus, a compact description of the horizontal geometry can be obtained by resorting to the parameters defining the analytic expression of each curve segment. Vertical geometry is described in terms of changes on the slope of a track.
In addition to the network topology and to the track geometry, a Rail Digital Map may include information about a variety of railway relevant assets, addressed in the following as railway infrastructure elements, that can be found on, under, over or next to the railway track, like balises, platform edges, rail signs, and traffic lights. Attributes of these elements, including their geometry, can be stored in tables characterized by unique element identifiers. The geometry of infrastructure elements whose location can be represented as a single point of a map, (i.e., boards, poles, etc.) will include the linear and the geographic coordinates of their representative point. In this case, the coordinates with respect to the linear reference system will include, in addition to the mileage, the across track distance from the track centerline and the height with respect to the rail plane. On the contrary, infrastructure elements for which the physical size or the size of the region of interest is not negligible cannot be represented as single points on the map. For those applications for which a simplified description of their geometry can be considered sufficient we can store in the Digital Map the coordinates of a bounding box representing the volume occupied by the infrastructure element. Depending on the considered element, the bounding box may degenerate into a rectangle or a segment.
Considering that the best practices for surveys aimed at building the network topology and track geometry have already been established since a long time, in this paper, we focus our contribution on the theoretical background and on the algorithms used to build the section of the Digital Map related to the railway infrastructure elements. More specifically, the methodology behind the design and generation of this component of the Digital Map is based on:
• The post-processing of imaging and ranging sensors' records (stereo cameras and LIDAR) aimed at detecting the railway infrastructure elements and estimating their relative position with respect to the train.
• The post-processing of the GNSS records aimed at detecting the presence of local phenomena (multipath, electromagnetic interferences, etc.) that may impair the estimation of the location of the train and estimating their temporal occurrence during a run.
• The combination of the train location derived from GNSS data with the object’s relative position and the absolute (geographic) object’s location based on previous runs stored in the Digital Map in order to set/update the absolute position of each infrastructure element. The process of building the Digital Map is incremental and the accuracy of the location of each Network Element improves by increasing the number of runs.
In this work, along with the full description of the railway digital map design, we focus on the computation of the railway infrastructure elements’ position for georeferencing purposes. More in detail, in this contribution we describe the procedures for computing the landmark coordinates through the use of on-board visual sensors and provide some simulation results.
References:
[1] https://voliera.eu/
[2] https://railgap.eu/
[3] https://www.railtopomodel.org/en/
[4] www.railml.org



Return to Session B4b