# TP-Lib: Train Positioning Library
[](https://github.com/matdata-eu/tp-lib/actions/workflows/ci.yml)
[](https://codecov.io/gh/matdata-eu/tp-lib)
[](https://crates.io/crates/tp-lib-core)
[](https://pypi.org/project/tp-lib/)
[](https://www.nuget.org/packages/TpLib/)
[](https://matdata-eu.github.io/tp-lib/)
[](LICENSE)
**Status**: Under construction
Train positioning library excels in post-processing the GNSS positions of your measurement train to achieve an unambiguous network location. This library is your map matching assistant specifically for railway.
## Features
- π **High Performance**: R-tree spatial indexing for O(log n) nearest-track search
- π **Accurate Projection**: Haversine distance and geodesic calculations with geo-rs
- π€οΈ **Train Path Calculation**: Probabilistic path calculation through rail networks using topology
- πΊοΈ **Interactive Path Review**: Browser-based map webapp to visually review and edit calculated paths before projection
- π **CRS Aware**: Explicit coordinate reference system handling (EPSG codes)
- β° **Timezone Support**: RFC3339 timestamps with explicit timezone offsets; naive (timezone-less) ISO 8601 datetimes are accepted on input and assumed to be in the host's **local** timezone. All emitted timestamps include an explicit timezone offset.
- π **Multiple Formats**: CSV and GeoJSON input/output
- π§ͺ **Well Tested**: 460 comprehensive tests (all passing) - unit, integration, contract, CLI, and doctests
- β‘ **Production Ready**: Full CLI interface with validation and error handling
- π **Automatic RINF Retrieval**: When a topology file is omitted, the library can download a bounding-box subset of the [ERA RINF](https://data-interop.era.europa.eu/) network on demand
## Train Path Calculation
The library supports continuous train path calculation using network topology (netrelations)
to determine the most probable route a train took through the railway network.
### How It Works
1. **Candidate Selection**: Find candidate track segments within cutoff distance of each GNSS position
2. **Emission Probability**: Calculate per-position probability using exponential decay of distance and heading alignment
3. **Viterbi Decoding**: Decode the globally optimal netelement sequence using a log-space Viterbi algorithm (HMM map matching, Newson & Krumm 2009). Transition probabilities are derived from shortest-path vs. great-circle distance through the topology graph. Bridge netelements are inserted automatically for path continuity.
### CLI Commands
```bash
# Default: Calculate path AND project coordinates onto it
tp-cli --gnss positions.csv --network network.geojson --output result.csv
# Calculate path only (export path segments without projection)
tp-cli calculate-path --gnss positions.csv --network network.geojson --output path.csv
# Simple projection (legacy mode, no path calculation)
tp-cli simple-projection --gnss positions.csv --network network.geojson --output projected.csv
# Use pre-calculated path for projection
tp-cli --gnss positions.csv --network network.geojson --train-path path.csv --output result.csv
# Review calculated path in browser before projection (integrated mode)
tp-cli --gnss positions.csv --network network.geojson --output result.csv --review
# Launch standalone webapp to review/edit a path file
tp-cli webapp --network network.geojson --train-path path.csv --output reviewed_path.csv
# Fetch the RINF topology covering a GNSS file (inspection helper).
# Writes the retrieved netelements + netrelations to GeoJSON even when
# validation reports issues (e.g. coarse geometries), then exits with the
# corresponding non-zero status code.
tp-cli fetch-topology --gnss positions.geojson --output topology.geojson
```
### Debug Output
Pass `--debug` to write intermediate HMM calculation results as GeoJSON files to a `debug/` subdirectory next to the output file.
See **[DEBUG.md](DEBUG.md)** for a full description of the debug output files, their properties, and a typical debugging workflow.
### Algorithm Parameters
| `--distance-scale` | 10.0 | Distance decay scale (meters) |
| `--heading-scale` | 2.0 | Heading decay scale (degrees) |
| `--cutoff-distance` | 500.0 | Max distance for candidate selection (meters) |
| `--heading-cutoff` | 10.0 | Max heading difference to accept (degrees) |
| `--probability-threshold` | 0.02 | Minimum emission probability for candidate inclusion |
| `--max-candidates` | 3 | Max candidates per GNSS position |
| `--resampling-distance` | None | Optional GNSS resampling distance (meters) |
### Performance
- **Benchmark**: 1000 positions Γ 50 netelements in ~900ΞΌs (target was <10s)
- **Scalability**: Sub-linear scaling with R-tree spatial indexing
- **Memory**: Handles 10,000+ positions efficiently
## Train Path Review Webapp
The library ships a companion browser-based map webapp (`tp-webapp`) that lets you visually inspect and edit a calculated train path before using it for projection. No npm, no Node.js, no frontend build step required β the web assets are embedded in the binary at compile time using `rust-embed`.
### Modes
| **Standalone** | `tp-cli webapp` | Review/edit an existing path file; save result to disk; server stays alive |
| **Integrated** | `tp-cli β¦ --review` | Pause the projection pipeline after path calculation; confirm or abort in the browser |
### What You Can Do in the Browser
- View all network segments on a Leaflet map (OpenStreetMap basemap, toggleable)
- Path segments are highlighted with a colour-coded confidence scale
- **Add** a netelement to the path by clicking it on the map β snapped to the correct topological position
- **Remove** a path segment by clicking it β segment reverts to the default style
- Sidebar shows an ordered list of segments with probability scores
- **Dark mode** toggle (auto-activates from OS `prefers-color-scheme`)
- **Close Tab** button for clean exit after the server shuts down
### Standalone Usage
```sh
tp-cli webapp \
--network network.geojson \
--train-path path.csv \
--output reviewed_path.csv
```
A browser opens at `http://127.0.0.1:8765`. Click **Save** to write the reviewed path; the server stays alive for further edits. Press Ctrl+C when done.
The saved file is accepted directly by `--train-path`:
```sh
tp-cli --gnss positions.csv --network network.geojson \
--train-path reviewed_path.csv --output result.csv
```
### Integrated Usage
```sh
tp-cli --gnss positions.csv --network network.geojson \
--output result.csv --review
```
1. CLI calculates the train path
2. Browser opens automatically β review and edit the path
3. Click **Confirm** β path artifact saved as `result-path.csv`; projection continues
4. Click **Abort** β CLI exits with non-zero code
### Build Without Web Server
```sh
cargo build --package tp-cli --no-default-features
```
## Project Structure
```
tp-lib/ # Rust workspace root
βββ tp-core/ # Core Rust library
β βββ src/
β β βββ models/ # Data models (GnssPosition, Netelement, ProjectedPosition, PathOrigin)
β β βββ projection/ # Projection algorithms (geom, spatial indexing)
β β βββ path/ # Train path calculation (candidate, probability, graph, viterbi)
β β βββ io/ # Input/output (CSV, GeoJSON, Arrow)
β β βββ crs/ # Coordinate reference system transformations
β β βββ temporal/ # Timezone handling utilities
β β βββ errors.rs # Error types
β βββ tests/
β β βββ unit/ # Unit tests
β β βββ integration/ # Integration tests
β βββ benches/ # Performance benchmarks
βββ tp-cli/ # Command-line interface
βββ tp-webapp/ # Interactive path review web server (axum + Leaflet.js)
βββ tp-py/ # Python bindings (PyO3)
βββ tp-net/ # .NET bindings (csbindgen + System.Text.Json)
```
## Quick Start
### Prerequisites
- Rust 1.91.1+ (install from [rustup.rs](https://rustup.rs/))
- Python 3.12+ (for Python bindings)
### Build from Source
```bash
# Clone repository
git clone https://github.com/matdata-eu/tp-lib
cd tp-lib
# Build all crates
cargo build --workspace
# Run tests
cargo test --workspace
# Run benchmarks
cargo bench --workspace
```
### Docker Usage
#### Production Deployment
Use Docker to run the CLI without installing Rust:
```bash
# Build production image
docker build -t tp-lib:latest .
# Run with mounted data directory
docker run --rm -v $(pwd)/data:/data tp-lib:latest \
--gnss-file /data/gnss.csv \
--crs EPSG:4326 \
--network-file /data/network.geojson \
--output-format csv > output.csv
# Or use docker-compose
docker-compose up tp-cli
```
#### Running Tests in Docker
Run the complete test suite including CRS transformation tests:
```bash
# Using docker-compose (recommended)
docker-compose run --rm test
# Or build and run test image directly
docker build -f Dockerfile.test -t tp-lib-test .
docker run --rm tp-lib-test
# Run specific tests
docker-compose run --rm test cargo test test_identity_transform
# Run only CRS transformation tests
docker-compose run --rm test cargo test crs_transform
# Interactive shell for debugging
docker run --rm -it tp-lib-test bash
```
**Why Docker for tests?**
- **Complete test coverage**: Runs all tests including CRS transformation tests
- **Consistent environment**: Same Rust version across all machines
- **No local setup needed**: No need to install Rust toolchain locally
- **CI/CD ready**: Use `Dockerfile.test` in GitHub Actions or other CI systems
### Usage Examples
```bash
# CLI usage - CSV input/output
tp-cli --gnss-file train_positions.csv \
--crs EPSG:4326 \
--network-file railway_network.geojson \
--output-format csv > projected.csv
# GeoJSON output with custom warning threshold
tp-cli --gnss-file positions.csv \
--crs EPSG:4326 \
--network-file network.geojson \
--output-format json \
--warning-threshold 100.0 > projected.geojson
# Custom CSV column names
tp-cli --gnss-file data.csv \
--crs EPSG:4326 \
--network-file network.geojson \
--lat-col lat --lon-col lon --time-col timestamp
```
### Library Usage
```rust
use tp_lib_core::{parse_gnss_csv, parse_network_geojson, RailwayNetwork};
use tp_lib_core::{project_gnss, ProjectionConfig};
fn main() -> Result<(), Box<dyn std::error::Error>> {
// Load railway network from GeoJSON
let netelements = parse_network_geojson("network.geojson")?;
let network = RailwayNetwork::new(netelements)?;
// Load GNSS positions from CSV
let positions = parse_gnss_csv(
"gnss.csv",
"EPSG:4326",
"latitude",
"longitude",
"timestamp"
)?;
// Project onto network with default config (50m warning threshold)
let config = ProjectionConfig::default();
let projected = project_gnss(&positions, &network, &config)?;
// Use results
for pos in projected {
println!(
"Position at {}m on netelement {} (accuracy: {:.2}m)",
pos.measure_meters,
pos.netelement_id,
pos.projection_distance_meters
);
}
Ok(())
}
```
## Automatic RINF Topology Retrieval
When you do not supply a `network` file, `tp-lib` can derive the bounding box from
your GNSS input (and optional path) and download a fresh subset of the ERA RINF
topology from a SPARQL endpoint.
- **Default endpoint**: `https://graph.data.era.europa.eu/repositories/rinf-plus`
- **Default buffer**: 1000 m around the GNSS extent
- **Default timeout**: 60 seconds per HTTP request
- **Coarse-geometry warning threshold**: netelements longer than 250 m without a WKT geometry
Outcome categories (mapped to typed errors / exit codes across all bindings):
| `invalid_gnss_input` | 4 | `TpLibInvalidGnssInputException` | `InvalidGnssInputError` |
| `rinf_missing_coverage` | 5 | `TpLibRinfMissingCoverageException` | `RinfMissingCoverageError` |
| `rinf_incomplete_topology` | 6 | `TpLibRinfIncompleteTopologyException` | `RinfIncompleteTopologyError` |
| `rinf_retrieval_failed` | 7 | `TpLibRinfRetrievalFailedException` | `RinfRetrievalFailedError` |
See per-language READMEs (`tp-cli/`, `tp-py/`, `tp-net/`) for end-to-end examples.
## Implementation Notes
### Performance
- **Target**: < 10 seconds for 1000 GNSS positions Γ 50 netelements (SC-001)
- **Memory**: Handles 10,000+ positions without exhaustion (SC-006)
- **Accuracy**: 95% of positions within 2m projection distance (GPS quality dependent)
- **R-tree Complexity**: O(log n) nearest-neighbor search
### Input Data Requirements
**GNSS CSV:**
```csv
latitude,longitude,timestamp,altitude,hdop
50.8503,4.3517,2025-12-09T14:30:00+01:00,100.0,2.0
```
- RFC3339 timestamps with timezone (e.g. `2025-12-09T14:30:00+01:00` or `2025-12-09T14:30:00Z`); naive ISO 8601 datetimes without timezone (e.g. `2025-12-09T14:30:00` or `2025-12-09 14:30:00`) are accepted on input and interpreted in the host's **local** timezone. All output timestamps are emitted in RFC3339 form with an explicit timezone offset.
- CRS must be specified via `--crs` flag
- Column names configurable with `--lat-col`, `--lon-col`, `--time-col`
**Railway Network GeoJSON:**
```json
{
"type": "FeatureCollection",
"features": [
{
"type": "Feature",
"properties": { "id": "NE001", "crs": "EPSG:4326" },
"geometry": {
"type": "LineString",
"coordinates": [
[4.35, 50.85],
[4.36, 50.86]
]
}
}
]
}
```
- LineString geometries (track centerlines)
- Unique `id` property per netelement
- `crs` property with EPSG code
### Troubleshooting
**"Large projection distance" warnings:****
Indicates GNSS position is far from nearest track (> threshold). Possible causes:
- GPS inaccuracy or poor signal quality
- Train on parallel track not in network
- Missing netelement in railway network
- CRS mismatch between GNSS and network
- Track geometry outdated or incorrect
Adjust threshold with `--warning-threshold` flag or investigate data quality.
**"No Python 3.x interpreter found" build error:**
Building with default features requires Python for PyO3 bindings. Disable with:
```bash
cargo build --no-default-features
```
Or install Python 3.12+ and ensure it's in your PATH.
### Known Issues
1. **Windows Build Dependencies**: Requires MSVC toolchain or mingw-w64 for some native dependencies
2. **Python Bindings**: Requires Python 3.12+ installed (excluded from tp-py crate builds by default)
## CRS Transformations
TP-Lib uses **proj4rs**, a pure Rust implementation of PROJ.4, for coordinate reference system transformations. This eliminates system dependencies and enables cross-platform compatibility.
**Key Features:**
- **Pure Rust**: No external C libraries required (libproj, sqlite3, etc.)
- **Zero system dependencies**: Works on Windows, Linux, macOS without installation
- **EPSG support**: Uses `crs-definitions` crate for EPSG code lookup
- **WASM compatible**: Can be used in browser environments
- **Always enabled**: CRS transformations are available by default
**Supported Transformations:**
TP-Lib has been tested with Belgian railway coordinate systems:
- EPSG:4326 (WGS84) β EPSG:31370 (Belgian Lambert 72)
- EPSG:4326 (WGS84) β EPSG:3812 (Belgian Lambert 2008)
- Any EPSG codes supported by [crs-definitions](https://docs.rs/crs-definitions/)
**Usage:**
```rust
use tp_lib_core::crs::CrsTransformer;
use geo::Point;
// Create transformer (EPSG codes or PROJ strings)
let transformer = CrsTransformer::new(
"EPSG:4326".to_string(),
"EPSG:31370".to_string()
)?;
// Transform point (automatic degree/radian conversion)
let wgs84_point = Point::new(4.3517, 50.8503);
let lambert_point = transformer.transform(wgs84_point)?;
```
**Technical Details:**
- proj4rs automatically handles radian/degree conversions for geographic CRS
- EPSG codes are resolved to PROJ strings using the crs-definitions crate
- Custom PROJ strings can be used directly instead of EPSG codes
- Transformation accuracy matches PROJ for standard 2D transformations
**Limitations:**
- proj4rs implements PROJ.4 API (2D transformations only)
- No 3D/4D or orthometric transformations
- Grid shift support is experimental
- For complex geodetic requirements, consider using [PROJ](https://proj.org/) directly
## Documentation
### API Documentation
**Online:** https://matdata-eu.github.io/tp-lib/
The documentation is automatically built and deployed on every push to `main`. It includes:
- **tp-core**: Core library API with examples
- **tp-cli**: Command-line interface documentation
- **tp-py**: Python bindings API reference
- **tp-net**: .NET bindings API reference
**Build locally:**
```bash
# Generate documentation for all workspace crates
cargo doc --no-deps --workspace
# Open in browser (on Windows)
start target/doc/index.html
# Open in browser (on Linux/macOS)
open target/doc/index.html # macOS
xdg-open target/doc/index.html # Linux
```
### Specification Documents
**Feature 001 β GNSS Projection**
- [Feature Specification](specs/001-gnss-projection/spec.md)
- [Implementation Plan](specs/001-gnss-projection/plan.md)
- [Data Model](specs/001-gnss-projection/data-model.md)
- [CLI Contract](specs/001-gnss-projection/contracts/cli.md)
- [Tasks](specs/001-gnss-projection/tasks.md)
**Feature 002 β Train Path Calculation**
- [Feature Specification](specs/002-train-path-calculation/spec.md)
- [Implementation Plan](specs/002-train-path-calculation/plan.md)
- [Data Model](specs/002-train-path-calculation/data-model.md)
- [Tasks](specs/002-train-path-calculation/tasks.md)
**Feature 003 β Train Path Review Webapp**
- [Feature Specification](specs/003-path-review-webapp/spec.md)
- [Implementation Plan](specs/003-path-review-webapp/plan.md)
- [Data Model](specs/003-path-review-webapp/data-model.md)
- [REST API Contract](specs/003-path-review-webapp/contracts/api.md)
- [CLI Contract](specs/003-path-review-webapp/contracts/cli.md)
- [Quickstart](specs/003-path-review-webapp/quickstart.md)
- [Tasks](specs/003-path-review-webapp/tasks.md)
### CI/CD & Workflows
This project uses automated workflows for continuous integration and deployment:
- π **Continuous Integration**: Automated testing, linting, and security checks on every push
- π¦ **crates.io Publishing**: Automatic release to Rust package registry
- π **PyPI Publishing**: Automatic release to Python package index
- π **Documentation Deployment**: Auto-deployed to GitHub Pages
See **[CI/CD Workflows Documentation](docs/WORKFLOWS.md)** for details on:
- Build and test automation
- Release process and version management
- Security and license validation
- Publishing to crates.io and PyPI
- Documentation deployment
### Constitution Compliance
This project follows the TP-Lib Constitution v1.1.0 principles:
- β
**I. Library-First**: Single unified library with quality external dependencies
- β
**II. CLI Mandatory**: Command-line interface for all functionality
- β
**III. High Performance**: Apache Arrow, R-tree spatial indexing
- β
**IV. TDD**: Test-driven development with FIRST TEST validation
- β
**V. Full Coverage**: Comprehensive test suite (unit, integration, property-based)
- β
**VI. Timezone Awareness**: DateTime<FixedOffset> for all timestamps
- β
**VII. CRS Explicit**: All coordinates include CRS specification
- β
**VIII. Error Handling**: Typed errors with thiserror, fail-fast validation
- β
**IX. Data Provenance**: Preserve original GNSS data, audit logging
- β
**X. Integration Flexibility**: Rust API + CLI + Python bindings + .NET bindings
## Contributing
This project follows strict TDD workflow:
1. Write test first (RED)
2. Implement minimum code to pass (GREEN)
3. Refactor while keeping tests green
See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines.
## License
Apache License 2.0 - See [LICENSE](LICENSE) for details
## Contact
TP-Lib Contributors - [GitHub Issues](https://github.com/Matdata-eu/tp-lib/issues)