Survey2CAD is being extended to rail: rail corridors captured with mobile mapping, rail-mounted scanners, or drones are turned into structured CAD geometry – on the same AI platform as for roads.
Rail Analysis in Preparation.
We are looking for surveying firms and rail partners who want to get involved early with their track data. Talk to us about a pilot project.
Same Platform, New Object Classes
Rail follows the same workflow as roads – from upload to CAD delivery.
Upload
Point cloud of the rail corridor from mobile mapping, rail-mounted scanner, or drone (LAS, LAZ, E57), including track sections spanning several kilometers. Specify coordinate system and height reference.
AI Detection
The AI detects the approved rail object classes (e.g., rails, masts, platform edges) in the point cloud and derives their position and geometry from them.
Completion & Review
Jaroona fills in occlusions and gaps (e.g., caused by vehicles or vegetation) and checks the result against the agreed quality criteria (Full-Service).
CAD Delivery
Structured geometry as DXF and 3D polylines, organized into layers by object class; other formats on request.
Why Railways Need This
Rail networks are being modernized corridor by corridor. Each corridor requires an accurate, up-to-date digital record – and today this is still largely traced by hand from the point cloud.
High effort per kilometer
A rail corridor often yields several million points per kilometer, up to gigabytes of data. Tracks, masts, switches, signals, and platform edges are still traced manually today – tying up specialist staff for weeks.
Short windows on the track
Surveys and measurements are only possible during possession windows, usually at night or on weekends. Evaluation must keep pace with the survey campaigns, otherwise raw data piles up.
Foundation for planning and maintenance
Expansion, digitalization, and maintenance require an accurate, up-to-date digital record per corridor in CAD (position, elevation, geometry of assets), which serves as the planning basis.
Survey2CAD automates this step – from point cloud to structured CAD geometry.
From Millions of Measurement Points to CAD Objects in the Track Corridor
A point cloud captures the track corridor in high detail – but a existing-condition plan does not need millions of points, only clearly defined elements: rails as lines, masts as locations with an axis, platform edges as continuous edges. Survey2CAD automates exactly this conversion.
Input
Point cloud from mobile mapping, rail-mounted scanner, or drone (LAS, LAZ, E57)
AI Processing
Classification, object recognition, vectorization, quality assurance
Result
DXF with 3D polylines in your layer structure; other formats on request
Automatic Detection of Railway Assets
Track systems have very regular geometry – and that is exactly what the AI takes advantage of. It combines machine learning with 3D geometry analysis and railway-specific rules, such as the spacing of the rails or the position of platform edges relative to the top of rail.
Analyzed Features
- Height relative to the top of rail
- Line continuity and track alignment
- Spacing and parallelism
- Vertical structures in the clearance envelope
- Intensity values (where available)
Object Classes We Are Developing for Rail
- Tracks (top of rail, track centerline)
- Switches and crossings
- Overhead line catenary masts
- Platform edges
- Signals and kilometer markers
- Track bed and surroundings
Rails
From the points of the rail heads, the rail top edges (ROT) on the left and right are derived as continuous 3D polylines. The track axis is created as the centerline between both ROT lines and serves as the reference for planning, chainage, and the existing asset.
Rail top edge left/right
Two continuous 3D polylines along each track, referenced to the highest point of the rail head. Gaps caused by occlusions are bridged and marked.
Track axis
Centerline between the two ROT lines as the reference for planning and chainage. Track spacing and gauge (standard gauge 1,435 mm) can be derived for plausibility checks.
Multi-track sections
Parallel tracks in station areas are detected separately and output individually as their own objects with a track ID.
Transitions
Bridges, level crossings, and tunnel portals are tracked continuously, so the track does not break off when the surroundings and point density change.
Switches and Crossings
Switches connect tracks and are particularly important in the existing-conditions plan. The AI detects the area where rails split (frog, switch blades, switch entrance), and derives the start, end, and diverging track from it. The rail top surfaces are carried through the switch continuously.
Switch Start and End
The switch entrance (start of the switch blade assembly) and switch end (end of the switch area) are output as 3D points, referenced to the track axis, with a switch ID.
Main Track and Diverging Track
Both tracks are output as separate rail top surface lines and track axes; the frog is marked as the branching point so the track topology remains traceable.
Crossings and Switch Sequences
Track crossings and consecutive switches (e.g., track connections and switch sequences in the station throat) are detected as a connected group and documented with switch connections. Complex cases are reviewed by Jaroona.
Overhead Line Masts
Masts are vertical structures placed at regular intervals along the track, usually to the side of the rail line, and in some cases as gantries or portals spanning several tracks. The AI recognizes them by their geometry (height, vertical alignment, recurring spacing along the route) and distinguishes them from vegetation and other structures; their location and axis are output.
Mast Base
Location as a 3D point at the transition to the foundation or ground, with mast ID and reference to the nearest track axis (lateral offset and height difference to the rail top).
Mast Axis
Vertical reference line from the mast base to the mast top; mast inclination and mast height can be derived from it.
Cantilever Arms
Position and height of the cantilever arms relative to the track; for gantries and portals spanning several tracks, also the assignment to the tracks being spanned.
Platforms
Platform edges are derived as continuous edges along the track, including their position and height relative to the top of rail. This is central to planning, conversion, and accessibility, particularly for step-free boarding.
Platform Edge
Continuous 3D polyline along the track-side edge of the platform. Where interrupted by stairs, structures, or occlusions, the edge is bridged and marked accordingly.
Height and Offset from Top of Rail
Platform height above the top of rail and lateral offset from the track centerline are determined along the edge and output as attributes, so deviations from the target height are easy to identify.
Platform Start and End
The start and end of each platform edge are captured as 3D points with platform and track IDs. This yields the usable platform length.
Signals, Kilometer Markers and Equipment
Point-like objects along the route are output as 3D locations with object class, ID and reference to the nearest track.
Signals
Main, distant and shunting signals as well as signal masts or brackets are captured as a location with side (left/right of the track), lateral distance to the track centerline and height, and assigned to the respective track. The signal type is only output where it can be reliably derived from the geometry; otherwise the assignment is made via the Jaroona check.
Kilometer Posts and Route Markers
Locations of kilometer posts, hectometer markers and other route markers are captured as points. They serve as reference points for referencing the kilometering along the route (details in the Route Reference section).
Additional Equipment
Extensible per project, e.g. cable ducts, switch cabinets, luminaires and warning signs. The object classes are defined jointly with the client and are only transferred to production after the quality threshold has been reached.
Trackbed and Surroundings
In addition to the track installations themselves, elements in the surroundings of the trackbed are also derived – just as with roads. They provide context for existing-conditions planning, drainage, clearance envelope, and maintenance.
Slope Crest and Toe
Slope crest and slope toe are captured as continuous 3D polylines on both sides of the track, separately for cuttings and embankments; the slope geometry serves as the basis for stability and drainage analyses.
Ditches and Drainage
Ditches, swales, and track drainage are identified from their terrain form and output as lines (ditch invert and edges); the flow direction can be derived from the elevation profile.
Noise Barriers
Wall crest and wall base are captured as 3D polylines with wall height above rail top and distance from the track centerline; interruptions (e.g., doors, openings) are marked.
Fences
The fence alignment is captured as a 3D polyline along the railway boundary, with interruptions at gates and crossings; fences in dense vegetation can be completed by Jaroona in the full-service option.
Railway Stations, Switch Approaches, Bridges, and Tunnels
Real track installations are complex: stations with many parallel tracks, switch approaches in the station throat, bridges and tunnels (with differing point density and sometimes poor position signal), and occlusions from parked vehicles, platform roofs, or vegetation. The AI therefore always considers the spatial context rather than viewing objects in isolation.
Contextual Awareness
Track alignment, neighboring objects, and plausibility feed into detection. A switch must connect to a main track, masts stand at regular intervals, and tracks run within plausible gauges and track spacings.
Flagging Uncertain Areas
Low-confidence areas, such as those under parked vehicles, in tunnels, or in dense vegetation, are highlighted and specifically reviewed by Jaroona before results are delivered.
Controlled Level of Automation
Uncertain results are never silently accepted. Every automatically generated geometry remains traceable and can be corrected in CAD by the surveyor.
Our Quality Rule
An object class – such as rail top surface, catenary mast, or platform edge – goes into production only once it reaches at least 90% AI maturity and meets the agreed quality criteria. Both conditions must apply.
AI maturity is the share of element length that can be accepted without rework. Example: if 10 km of rail top surface was detected, at least 9 km must be usable without manual correction for the class to reach the 90% threshold.
In Development
The object class is being trained and internally tested. It is not yet released for customer projects.
Clearly Labeled Test
Used only in pilot projects. Results are explicitly marked as a test, and the customer is aware of the maturity level.
Production
From at least 90% AI maturity and fulfilled quality criteria. Delivered under the regular full-service model.
This creates transparency: for every object class, customers know where they stand and what they can expect.
Human in the Loop
Our goal is not to replace the surveyor. The AI handles the time-intensive digitization – tracing rails, masts, platform edges, and other objects. Professional responsibility and final approval remain with people.
AI processes the point cloud
The AI recognizes the approved object classes across the entire track corridor and generates geometry from them – without anyone having to trace point by point.
Quality is evaluated
A reliability score is calculated for each detected object, based on geometry, context, and plausibility checks, such as track gauge or mast spacing.
Uncertain areas are flagged
Areas with low reliability – for example under parked vehicles, in switch zones, or where vegetation is present – are highlighted so that reviewers can focus their attention exactly there.
Expert review and approval
Qualified specialists review the flagged areas, correct them where needed, and approve the result before it is delivered. In the Full-Service, Jaroona handles this review; alternatively, your own surveying office can do it.
Output Formats
Results are delivered in formats that integrate directly into existing CAD, GIS, and BIM workflows.
Standard
DXF with 3D polylines for SOK left and right, track axis, platform edges, embankments, and other linear objects, plus 3D points for masts, signals, and kilometer markers. The data is organized into layers by object class; the coordinate system and height reference follow the client's specifications.
On Request
- CAD/BIM: DWG, IFC
- GIS: GeoJSON, SHP, GeoPackage
- Point cloud: classified point cloud (LAS/LAZ)
- Documentation: QA reports with the AI maturity per object class and the reviewed areas
Layer names, attributes, and data structure are adapted to your CAD standards, such as existing layer conventions or templates. Feel free to discuss this with us during the pilot project.
Track Referencing and Kilometer Positioning
Detected objects are referenced along the track. This allows every rail, mast, and signal to be assigned to a track section and a fixed position, rather than existing only as coordinates in the point cloud.
Kilometer Positioning
Each object is assigned a chainage along the track centerline. The basis is reference points such as kilometer markers or reference points supplied by the client; the accuracy depends on the quality of these reference points.
Section ID
Track sections such as line number, track, or station/block section are written as IDs to the objects. Evaluations by section are thus directly possible.
Attributes for Assets and Maintenance
Objects can be assigned attributes such as object type, position left or right of the track, capture date, and inspection status. These form the basis for asset management and maintenance planning; the attribute schema is agreed upon with the client.
Typical Applications
Corridor Renovation
current inventory as the planning basis
As-Built Plans
CAD plans from the survey instead of manual tracing
Planning and Modification
e.g., platforms, switches, overhead catenary
Maintenance
structured objects instead of unstructured point clouds
Recurring Surveys
new measurement campaigns using the same processing chain
Cloud by Default – On-Premise on Request
Survey2CAD runs by default as a cloud service – without any GPU infrastructure on the customer's side. Private cloud, customer-owned cloud, on-premise, and hybrid models are available on request.
Data Privacy and Data Sovereignty
Point clouds of rail infrastructure contain sensitive infrastructure information. Processing takes place on servers in the EU; data transmission is encrypted, project workspaces are separated, and access rights are role-based. Deletion after project completion upon request.
Rail Pilot Projects
We are looking for surveying firms and partners from the rail infrastructure sector who would like to test Survey2CAD on their own track data – with a clearly defined pilot scope.
What You Bring
One to two track sections as a point cloud (LAS, LAZ, or E57), e.g. from mobile mapping, a rail-mounted scanner, or a drone, including the coordinate system and height reference.
What You Receive
CAD output (DXF with 3D polylines) for the agreed object classes, completed and reviewed by Jaroona – labeled as pilot results, with the maturity level stated for each object class.
How It Works
Initial discussion on data situation and goals, data handover, evaluation, joint review of the results, and feedback on quality and desired additional object classes.
Pilot partners actively shape the further development of rail object classes and gain early experience with the method.
Survey2CAD
AI that turns point clouds into finished CAD plans.
© 2026 Jaroona. All rights reserved.
Point Cloud to CAD for Rail Infrastructure