Real-Time Embedded SystemsCycle II · 2026-1
Real-time embedded speed enforcement with computer vision
An embedded system that watches a real street in Ayacucho from an IP camera, measures the speed of every vehicle, and files an automatic infraction report when someone crosses the limit.
Speeding is the leading cause of road incidents in Peru, behind 26.6% of the crashes recorded by the ONSV in 2023. Traditional traffic counting — manual, or inductive loops in the pavement — is expensive, invasive, and tells you nothing about vehicle type or speed. This project reuses infrastructure that is already installed: an ordinary surveillance camera.
From the video of an IMOU Cruiser SE+ IP camera mounted 9 m above Jr. Carlos F. Vivanco in Ayacucho, a YOLO-family CNN detects and classifies vehicles (car, motorcycle, minibus), ByteTrack assigns persistent identities, and a calibration homography converts inter-frame displacement into instantaneous per-vehicle speed. The system measures three physical traffic variables — flow by line counting, per-vehicle speed, and density inside a region of interest — and overlays them on a single interface, raising an alarm on speeding or congestion.
When a vehicle passes the 30 km/h limit, the system marks the offender and triggers automatic licence-plate recognition on an NVIDIA Jetson Orin NX. Each reading is transmitted, HMAC-signed, to an ingest API that cross-checks it in parallel against the Peruvian official sources SUNARP, SOAT (APESEG) and MTC, generating an infraction report with the plate, the recorded speed and the event timestamp. Field tests, daytime and nighttime, recorded 20–22 FPS with 18–31 ms of inference per frame. The methodological groundwork comes from RICUY, the ADAS built by the same group.