Datasets:

Kheyro
/

cyclist-intention-2d-keypoints

license: cc-by-4.0
task_categories:
  - other
tags:
  - cyclist
  - vulnerable-road-users
  - pose-estimation
  - behavior-prediction
  - traffic-safety
pretty_name: Cyclist Behavior 2D Keypoints V5
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*.parquet
      - split: validation
        path: data/validation-*.parquet
      - split: test
        path: data/test-*.parquet

Cyclist Behavior 2D Keypoints V5

Version 2.0.0 is a complete replacement of the earlier intersection release. It contains de-identified cyclist skeleton scene clips for bus-stop/road behavior prediction.

Summary

32 participants, 559 scene clips, 99,734 frames
33 two-dimensional pose landmarks (66 coordinates)
Stage1 labels: head look, upper-limb rotation, and pedaling intervals
Stage2 labels: straight, yield, and overtake
canonical Fold 1 participant-disjoint train/validation/test split
all five participant folds in participant_folds.parquet

Split	Sequences	Frames
Train	369	65,235
Validation	103	18,491
Test	87	16,008

annotations.parquet contains sanitized Stage1 gesture intervals and exclusion flags. sequences.csv contains clip-level maneuver labels, route anchors, FPS, and Stage2 eligibility.

Privacy

Raw video, original filenames, calendar dates, session IDs, direct participant numbers, and source paths are excluded. Release-scoped HMAC identifiers preserve grouping without publishing the source mapping. Pose trajectories can retain behavioral signatures; users must not attempt re-identification or linkage with external identity data.

Intended use and limitations

Intended for cyclist behavior, temporal pose, and participant-independent evaluation research. The simulator domain and variable source frame rate (12.690–19.423 FPS) limit direct real-world generalization. This dataset must not be used for biometric identification, surveillance, or safety-critical control without independent validation.

License

CC BY 4.0. Cite the dataset and indicate modifications.