Has anyone used citywide connected-vehicle or video-derived trajectory datasets as near-miss proxies to prioritize signal retiming and crossing safety? I’m compiling a lightweight toolkit and need downloadable trajectories at about 10 Hz for at least 48 hours per site to validate conflict rate and queue spillback models. Happy to trade back a Python notebook that estimates spillback risk and delta delay across three timing plans; it cut peak delay 14% on NE 45th last May without degrading pedestrian clearance.
Citywide CV data: pull Basic Safety Messages (10 Hz) from the ITS Public Data Hub (ITS Joint Program Office | ITS DataHub) and geofence your signals to aggregate >48 h per site for conflict and spillback checks. Video-derived sets like inD/rounD are clean but won’t hit your “10 Hz for at least 48 hours per site” requirement; sanity-check GNSS quality and apply jitter filters before stitching trajectories.
Building on @amanda_brown91, use the NYC or Tampa CV Pilot data on the ITS DataHub (ITS Joint Program Office | ITS DataHub) and join BSM to SPaT/MAP per RSU so you can lane-match and compute PET/TTC plus queue dwell by movement. Do you need ped/bike conflicts or strictly vehicle-vehicle? Either way, run a simple Kalman smoother on headings/positions because GPS wobble can make 10 Hz look like espresso jitters.
Map-match the BSMs to MAP lanes with a probabilistic matcher to tame GPS wobble, then compute TTC/PET only within the same movement; GraphHopper’s map-matching works well: GitHub - graphhopper/map-matching: The map matching functionality is now located in the main repository https://github.com/graphhopper/graphhopper#map-matching. For threshold tuning before scaling to the long CV runs, the inD intersection set has clean tracks (25 Hz) even if it’s shorter than 48 h. Are you targeting peds too, or just vehicle-vehicle conflicts?