theiatron

Traffic lights in cities like Lagos mostly run fixed timers: they give green to empty roads while queues build on busy ones. Adaptive signals fix this, but deciding whether a junction would benefit normally needs a site survey, manual traffic counts and a modelling consultant. Most junctions never get that study. Theiatron Junction Report answers "would smart signals help this junction?" from one video. Upload a clip of any junction and in about two minutes you get: 1. MAP: SAM 3.1 segments the road surface, vehicles, people and road markings. Road under parked or moving vehicles is filled in, and marking directions recover the camera perspective. No hand tracing, no survey. 2. COUNT: vehicles are tracked, approaches are found from their paths, and each vehicle is counted crossing its approach's line, clipped to the SAM road. You get vehicles per hour on each approach. 3. SIMULATE: the measured traffic is replayed in the SUMO traffic simulator under a generic fixed timer and under Theiatron's adaptive controller, on identical arrivals over five random seeds, with a side-by-side animation. Example: 4 approaches, 19 vehicles counted; simulated waiting fell from 58.7 s to 11.0 s per vehicle against a generic untuned timer, better on 5/5 seeds. Against a tuned fixed timer our benchmark shows 15.1% lower delay (held-out seeds, simulation). Every report labels what is measured, estimated and simulated; these are not field results. Live: https://web-production-37892.up.railway.app Theiatron's controller and simulator benchmark existed before the hackathon; the SAM 3.1 junction mapping and the hosted report were built during it, in separate commits.