Banshee

Measure velocity, not identity.

Banshee publishes open street-motion datasets from fixed roadside sensors, with an emphasis on reproducible traffic-safety and perception research. Our work combines radar speed measurement with replayable LiDAR captures and derived street scenes.

We publish raw observations alongside derived data where practical so that results can be reproduced, challenged, and improved as the processing pipeline evolves. Our systems do not use cameras or intentionally collect faces, licence plates, or other identifying fields.

Datasets & tools

• San Francisco street speeds provides LiDAR captures and derived street scenes for replay, analysis, and research. Manifests record capture intervals, geographic identifiers, file digests, and provenance.
• Source code contains the open-source radar and LiDAR processing pipelines, their maths, and current research work.
• velocity.report is the project site and local-first street-speed monitoring tool. Radar observations remain on the operator's device and can be turned into reports for understanding street speeds.

Research uses

The datasets are intended to support research and experimentation in areas including object tracking, trajectory reconstruction, speed estimation, traffic behaviour, sensor processing, street-scene reconstruction, and reproducible evaluation of roadside perception pipelines.

Individual dataset cards document the available formats, collection methodology, spatial and temporal organization, provenance, and limitations.

Data principles

We favour measurements that answer useful street-safety questions while minimizing the collection of identifying information. Raw sensor observations are treated as the durable research artifact; derived scenes, tracks, metrics, and interpretations may evolve as the processing pipeline improves.

Raw point clouds capture activity in public space. Although the datasets contain no camera imagery or intentionally identifying fields, users should treat raw sensor data as potentially sensitive and review each dataset card for collection, privacy, licensing, and redistribution guidance.

Research status

The radar pipeline is operational and used for street-speed measurement. LiDAR processing remains active research. Publishing replayable source data alongside derived outputs allows new methods and pipeline changes to be evaluated against the same observations.