OjoVial

Inspiration: At intersections like San José’s Circunvalación, city cameras work in isolation, and following a vehicle across them means manual review. What I built: OjoVial, a physical dual-camera scale testbed and web dashboard for prototyping cross-camera vehicle tracking without access to municipal feeds. How it works: Muse Spark 1.3 names candidate objects from both cameras. SAM 3.1 tracks the chosen one through each video from a text prompt. Muse Spark 1.3 then compares crops from both views and returns a verdict, confidence and visual evidence. Result: A working tabletop proof of concept chaining Meta’s models end to end. Each verdict is a model judgment on one moment, and the method hasn’t yet been validated on real traffic footage.