Bellevue 116th NE12th
Signalised intersection with wide-angle surveillance view.
About the research
Traffic Node Video Dataset 2.0 extends the previously published Traffic Node Video Dataset and preserves its original purpose: providing reusable video and trajectory data for traffic analysis, modelling and movement prediction. The new release updates the complete processing chain and introduces a reproducible map-based correction workflow.
The dataset contains recordings from eight traffic scenes, including signalised intersections, roundabouts and a pedestrian-oriented urban plaza. Object detections, tracked trajectories, cleaned trajectories, transformation data and software examples are published together so that the processing stages can be inspected and repeated.
Release highlights
8 traffic scenesYOLO11xByteTrackParquetHomographyJupyter
- Video material reprocessed with the YOLO11x detector.
- Previous DeepSORT stage replaced by the Ultralytics tracking workflow using ByteTrack.
- Per-video detection tables and structured trajectory exports.
- Filtering, filling and smoothing methods for trajectory refinement.
- Map-based geometric correction and isotropic coordinate normalisation.
What changed in version 2.0?
| Component | Previous release | Version 2.0 |
|---|---|---|
| Object detection | YOLOv7-based detection pipeline. | YOLO11x-based reprocessing with updated Ultralytics inference. |
| Object tracking | DeepSORT-based trajectory generation. | Ultralytics tracking workflow configured with ByteTrack. |
| Data storage | SQL and pickle-oriented trajectory exports. | Per-video Parquet detection tables and structured trajectory packages. |
| Trajectory preparation | Filtering, filling and ironing utilities. | Extended filtering, interpolation, smoothing, outlier analysis and reproducible examples. |
| Geometry | Image-plane coordinates. | Image-to-map homography, inverse projection, reprojection error and isotropic normalisation. |
| Reproducibility | Introductory examples and notebooks. | Expanded scripts, notebooks, sample data and a documented dataset package. |
Quantitative detector and tracker benchmark values will be reported in the associated publication after validation on the same annotated evaluation subset.
Processing pipeline
Video inputYOLO11x detectionByteTrack trackingTrajectory exportFiltering + fillingMap correctionAnalysis + publication
Featured datasets
Each scene is represented by its original video material and the corresponding detection and trajectory records. The preview images below are served directly from this repository.
Bellevue 150th Eastgate
Complex signalised junction with multiple turning movements.
Bellevue 150th Newport
Urban intersection used in trajectory clustering experiments.
Bellevue 150th SE38th
Multi-lane intersection suitable for lane-level analysis.
Bellevue Bellevue NE8th
Dense downtown traffic with diverse vehicle movements.
City of Orange Plaza
Urban plaza with mixed vehicle and pedestrian activity.
Historic Jonesborough
Small-town street environment with different scale and perspective.
Polanow roundabout
Roundabout scene used for detection and tracking demonstrations.
Software framework
Dataset loader
Loads manifests, per-video detections and trajectory records for Python-based analysis.
Trajectory visualisation
Plots raw, filtered, filled, smoothed and geometrically corrected trajectories.
Filtering and outlier analysis
Removes short, low-motion or application-specific trajectories and inspects OPTICS outliers.
Filling and smoothing
Interpolates missing detections and stabilises position, bounding-box and motion features.
Homography transformation
Estimates image-to-map transformation and calculates inverse projection and reprojection error.
Isotropic normalisation
Applies one shared spatial scale to preserve the relative geometry of the coordinate axes.
Sample code and notebooks
The repository contains executable examples for dataset loading, trajectory visualisation, filtering, filling, smoothing, homography estimation and normalisation.
Future Transportation HG-SMG-TC revision
Homography-guided semantic maneuver graph trajectory clustering
A manuscript-specific public reproducibility package is available for the revised Future Transportation study on five Bellevue intersections. It includes frozen protocols, reproducibility code, compact result tables, final evidence figures, reference-label documentation, homography quality summaries and OneDrive staging documentation for large redistributable artifacts.
Dataset access
Storage note. The full-resolution videos and generated trajectory packages are hosted on OneDrive because their size exceeds the practical limits of the GitHub repository. The public folder remains stable while individual files are updated.
| Package | Published content |
|---|---|
| 00_README_METADATA | Dataset manifest, scene metadata and processing configuration. |
| 01_ORIGINAL_VIDEO_REFERENCES | Original traffic videos and source references. |
| 02_YOLOV11X_DETECTIONS | YOLO11x detection data stored as per-video Parquet files. |
| 03_TRACKED_TRAJECTORIES | Tracked trajectory records and the trajectories.joblib export. |
| 04_CLEANED_TRAJECTORIES | Filtered, interpolated and smoothed trajectory variants. |
| 05_MAP_BASED_GEOMETRIC_CORRECTION | Control points, transformation matrices and corrected trajectory outputs. |
| 06_PERSPECTIVE_TRANSFORM_TOOL | Perspective transformation tool, examples and configuration. |
| 07_RESULTS_AND_STATISTICS | Generated summary tables, figures and validation outputs. |
| 08_FIGURES_FOR_PAPER_AND_PRESENTATION | Publication and presentation figures. |
| 10_SOFTWARE_TOOLS | Supporting scripts and software packages. |
Related publications
Suggested citation
Agg, A. D., Horváth, A.: Traffic Node Video Dataset 2.0: Enhanced Vehicle Trajectory Extraction with YOLO11x and Map-Based Geometric Correction. Tavaszi Szél Conference, 2026.
Final bibliographic details will be added after publication.