PEM-Flow: A trajectory dataset of controlled pedestrian and e-bike mixed flow with motor-vehicle probes

PEM-Flow provides trajectories from 18 controlled experiments of pedestrian–e-bike mixed flow with motor-vehicle probes. The dataset covers 19 video segments and 132,663 source frames, with 7,143,147 automatically processed records and 6,887,738 records after human review. The reviewed interaction region contains 2,819,547 records. This deposit contains the complete automatic and human-reviewed trajectory and motion tables, row-provenance tables, trajectory plots, downstream metrics, 870 reference images, human revision records, image observations, stabilization matrices, and reference-line measurement inputs and height-comparison results. The processing code, detector model, configurations, annotation XML, executable human revision rules, and three 10-second demonstration videos with matching outputs are available in the PEM-Flow GitHub repository. To use the code and data together, clone the GitHub repository with Git LFS, extract PEMFlow_data.zip, and copy its PEM-Flow/code/ and PEM-Flow/data/ directories into the cloned PEM-Flow directory, merging matching directories. Follow code/README.md or code/README_zh.md for installation and calculation commands. The supplied formal results are in data/automatic/, data/revised/ and data/downstream/. Data are licensed under CC BY 4.0. Software, third-party components and the model retain their respective licences in the GitHub repository.